Primary-secondary unmanned aerial vehicle cluster scheduling method and system

Abstract

The invention discloses a primary and secondary unmanned aerial vehicle cluster scheduling method and system, and relates to the technical field of unmanned aerial vehicle cluster cooperative control. The invention aims to solve the problems of single control architecture, unreasonable resource allocation, unstable coordination and high missing rate of the existing master-slave system. The core of the technical scheme is to construct a three-layer framework of'an air mother machine-two types of son machines-ground terminals', schedule AI crossing machines / high-speed crossing machines according to a preset multiple of a target number, dynamically distribute tasks in combination with a target priority, and support switching between'single-rack accurate throwing 'and'multi-rack batch launching'; cooperative interaction between the master machine and the slave machine is realized through MESH networking and magnetic attraction interface double links, and a whole-process closed loop of target detection, instruction issuing, state returning and missing network target blocking is formed. The task coverage efficiency is improved by more than two times, the response delay is smaller than or equal to 100 ms, the missed interception rate is smaller than or equal to 5%, the resource waste rate is reduced by 60%, and the method can be widely applied to scenes such as defense interception and safety prevention and control and meets diversified task requirements.

Description

Technical Field

[0001] This application relates to the field of collaborative control technology for unmanned aerial vehicle (UAV) swarms, and particularly to the scheduling of multiple types of sub-machines under a hierarchical architecture. Background Technology

[0002] With the rapid development of UAV technology, mother-daughter UAV systems have been widely used in multi-target mission scenarios due to their "mother-unit coordination + daughter-unit collaboration" operational mode. However, existing mother-daughter UAV systems still have many technical shortcomings:

[0003] 1. Simple control architecture: Most systems adopt a "single-level control" mode, where the master machine only undertakes the function of mounting and transporting slave machines. It lacks the ability to schedule multiple types of slave machines in a hierarchical manner, resulting in a limited task coverage and an inability to cope with distributed, large-scale target tasks.

[0004] 2. Unreasonable resource allocation: The sub-unit types are poorly adapted to the target requirements. There is no dynamic allocation mechanism based on the target priority and the performance differences of the sub-units. This easily leads to resource waste problems such as "high-speed sub-units performing low-speed tasks" and "precision sub-units performing assault tasks". Furthermore, it does not support flexible switching between "single precision drop" and "multiple batch launches".

[0005] 3. Unstable collaborative interaction: The master and slave machines mostly rely on a single communication link (wireless or wired), which is prone to signal interruption in complex electromagnetic environments or long-distance scenarios. Furthermore, a complete closed loop of "target detection - command issuance - status feedback - interception of missed targets" has not been formed, resulting in high task response delay and high miss rate.

[0006] 4. Insufficient adaptability: Existing systems are mostly designed for a single type of submachine, and cannot simultaneously meet the scheduling needs of low-to-medium speed precision AI racing drones and high-speed assault racing drones, making it difficult to cope with diverse mission scenarios.

[0007] Therefore, there is an urgent need for a parent-child drone swarm scheduling solution with a hierarchical collaborative architecture, dynamic resource allocation, stable interaction links, and a closed-loop process to solve the problems of resource waste, response delay, and high miss rate in existing technologies. Summary of the Invention

[0008] (a) Technical issues

[0009] This invention aims to address the core technical deficiencies exposed in the practical application of existing mother-daughter unmanned aerial vehicle (UAV) systems. These deficiencies are interconnected and lead to low overall system performance, as detailed below:

[0010] 1. Poor compatibility between various types of submachines and multiple target tasks, and a rigid resource allocation mechanism leading to serious waste.

[0011] Existing mother-daughter unmanned aerial vehicle (UAV) systems lack a precise matching logic between "daughter performance and target requirements." Resource allocation often adopts a "random allocation" or "fixed allocation" mode, failing to consider the compatibility between target priority, motion characteristics, and differences in daughter performance, resulting in significant resource waste.

[0012] The existing system lacks a dynamic adjustment mechanism: when the number or priority of targets changes, it cannot adjust the allocation of sub-machines in real time.

[0013] The configuration scheme has the following problems: when a new high-priority target is added, it is impossible to schedule idle high-speed submachines in time, which leads to the target escaping; in addition, the submachine type ratio is fixed, and most systems have an insufficient proportion of high-speed submachines, which cannot cope with clustered high-priority targets, while there is too much redundancy of medium and low-speed submachines, which further aggravates the waste of resources.

[0014] 2. The single-link communication architecture has weak anti-interference capabilities, and unstable collaborative interaction leads to task response delays.

[0015] Existing mother-daughter UAV systems mostly use a single wireless link for communication, without considering signal interference and transmission attenuation issues in complex scenarios. In complex environments such as electromagnetic interference and terrain obstruction, a single wireless link is prone to signal packet loss or even interruption, leading to increased command delivery delays and failing to meet real-time scheduling requirements.

[0016] The existing system has two major communication defects: First, it lacks data transmission priority classification, and core instruction data and non-critical status data are transmitted together, causing the bandwidth of instructions to be squeezed out by non-critical data, and the transmission delay is further increased; Second, there is no link backup mechanism. When the wireless link is interrupted, the slave unit cannot receive subsequent instructions and can only execute according to the preset program. It cannot cope with dynamic changes in the target, which ultimately leads to the deviation of task execution.

[0017] 3. The task process is open and lacks a closed loop; the absence of a mechanism to catch missed targets leads to a persistently high miss rate.

[0018] Existing mother-daughter UAV systems often only go as far as "daughter launch / drop - target interception attempt," failing to form a complete closed loop of "target detection - execution - feedback - re-interception," resulting in missed targets not being handled in a timely manner. The missed interception problem in existing systems mainly stems from three scenarios: First, targets are not tracked: due to insufficient number of daughter drones or unreasonable allocation, some targets are not covered by any daughter drone, and the system lacks a target coverage detection mechanism, leading to missed interception. Second, daughter drone mission failure: after a daughter drone fails to intercept due to insufficient power, target loss, or equipment failure, the system does not trigger an alternative plan, allowing the target to escape directly. Third, newly added targets are missed: during mission execution, newly added unassigned targets lack real-time detection and rapid scheduling mechanisms, resulting in them not being intercepted. Furthermore, existing systems lack priority scheduling logic for re-intercepting daughter drones. When multiple missed targets appear simultaneously, daughter drones are randomly scheduled, causing high-priority missed targets to not be handled first, further exacerbating the missed interception problem.

[0019] 4. The control architecture is singular and rigid, unable to achieve hierarchical collaboration among multiple types of submachines, and the scope of task coverage is limited.

[0020] Existing mother-daughter UAV systems mostly adopt a two-tier control architecture of "mother-daughter". The mother unit only undertakes the functions of mounting the daughter units and forwarding simple commands, lacking the overall scheduling capability of an "airborne hub" and unable to manage multiple types of daughter units hierarchically. At the same time, the two-tier architecture causes the ground terminal to directly face all daughter units, resulting in excessively long command transmission paths. Moreover, the ground terminal's computing power is limited and cannot handle the status data and scheduling requirements of a large number of daughter units. When the number of daughter units is large, the command generation delay increases, making real-time scheduling impossible and causing some daughter units to respond lagging in multi-target missions.

[0021] 5. The deployment mode of the sub-machines is limited and cannot adapt to different target distribution scenarios.

[0022] Existing systems have fixed deployment modes for their sub-units, mostly supporting only "sequential launch / drop of individual units," and cannot switch deployment modes based on target distribution. For clustered targets, sequential deployment results in excessive time differences in the arrival time of sub-units at the target area, by which time the targets have dispersed, significantly reducing interception efficiency. Conversely, for dispersed targets, batch deployment leads to excessive flight distances for the sub-units.

[0023] The battery drains too quickly, making it unable to complete the interception task.

[0024] (II) Technical Solution

[0025] This invention provides a method and system for scheduling a cluster of mother and child unmanned aerial vehicles (UAVs). Through an integrated design of "three-layer architecture + dynamic allocation + dual-link collaboration + closed-loop scheduling," it fundamentally solves the shortcomings of existing technologies. The specific technical solution is as follows:

[0026] 1. Scheduling method for mother-daughter UAV clusters

[0027] This method, centered on "layered collaboration, dynamic adaptation, stable interaction, and closed-loop controllability," achieves efficient execution of multi-objective tasks through the following four key modules:

[0028] (1) In-depth design of the three-tier collaborative architecture

[0029] The three-tier architecture is not a simple hierarchical division, but a functional collaboration system based on task flow. Each tier interacts through standardized interfaces and protocols to ensure complementary functions and data interoperability. The overall system architecture diagram is shown in Figure 2.

[0030] ①Airborne Mothership: Airborne Hub and Collaborative Core

[0031] Using a heavy-duty rotorcraft as its platform, it integrates three core functions: target detection, relay communication, and sub-aircraft control, serving as an airborne hub for sub-aircraft deployment and collaborative interaction.

[0032] Core functions: It has the ability to mount multiple AI racing drones stably; it integrates a target initial detection module and an AI recognition unit, which can scan and identify target areas and extract key target information; it is equipped with a relay communication module, which supports long-distance two-way data interaction with ground terminals and drones, and has wind and interference resistance capabilities, making it suitable for complex operating environments;

[0033] Key function: Coordinate the deployment of slave units, relay target information to ground terminals, receive instructions from ground terminals and send them to slave units, realizing information relay between "ground-master unit-slave unit".

[0034] ② Two types of sub-machines: Functional differentiation design and adaptation

[0035] To address different mission requirements, two types of AI-powered racing drones were designed: a low-to-medium speed precision type and a high-speed assault type. Both types of drones adopt a modular design, supporting rapid replacement and upgrades.

[0036] Medium- and low-speed precision AI racing drone: It has target recognition and guidance capabilities, its flight speed is adapted to the needs of precision operations, and its endurance meets the requirements of mission execution; equipped with an image acquisition module and AI recognition algorithm, it can start precise terminal recognition and guidance within a preset distance of approaching the target, and is suitable for the precise detection and interception of medium- and low-priority targets and static / low-speed moving targets.

[0037] High-speed assault racing aircraft: It adopts a streamlined structural design, has the ability to fly at high speed and strike quickly, and its endurance meets the requirements of short-distance high-speed operations; it is equipped with a multi-mode guidance module, which can activate terminal guidance within a preset distance of approaching the target, and is suitable for rapid attack and interception of high-priority targets and high-speed moving targets.

[0038] Common design: Both types of sub-units are equipped with standardized magnetic interfaces, which are physically compatible with the interfaces of the mother unit's throwing unit and the launcher, respectively, supporting rapid docking, data pre-transmission and start signal triggering.

[0039] ③ Ground terminal: Integrated control and interaction core

[0040] It consists of a transmitter, operating system terminal, domestically produced interactive terminal, and dedicated control software, forming an integrated "hardware + software" management and control system.

[0041] Launcher rack: It has multiple independent slots, each slot integrating a magnetic interface, power supply interface and guiding mechanism, supporting the storage, automatic charging and discharging and vertical launch of high-speed slave units; it integrates an environmental monitoring unit and a safety protection module, which can monitor environmental parameters such as temperature, humidity and abnormal smoke, and automatically cut off the power and alarm when abnormalities occur, realizing the integration of "storage, transportation and launch";

[0042] Operating system terminal: It has protocol conversion, data forwarding and command processing functions. It is connected to the transmitter via wired connection and supports communication adaptation with interactive terminals, master units and slave units to ensure stable transmission of commands and data.

[0043] Dedicated control software: possesses four core functions: equipment management, target operation, command issuance, and data feedback.

[0044] Equipment Management: Real-time display of the operating status of the main unit, slave unit, and transmitter; timely triggering of alarms when equipment connection is abnormal;

[0045] Target operation: Synchronizes real-time images from the master machine, supports target annotation and priority setting, and automatically generates target files containing target coordinates and type;

[0046] Command issuance: Based on the target requirements and the performance of the slave unit, a scheduling command is generated and forwarded to the master unit or transmitter via the terminal. The slave unit receives the command and sends back a confirmation signal.

[0047] Data feedback: Real-time display of the flight trajectory and target recognition results of the sub-unit, automatic storage of mission logs, and support for exporting review data.

[0048] (2) Dynamic allocation logic of slave machine resources: precise adaptation and flexible scheduling

[0049] Based on a three-dimensional allocation model of "target requirements - sub-machine performance - scenario adaptation", the optimal utilization of sub-machine resources is achieved. The specific process is as follows:

[0050] ① Target information collection and priority determination

[0051] Data collected: The AI ​​recognition module of the airborne mothership scans the target area and collects key information such as the number, coordinates, motion status, and type characteristics of the targets;

[0052] Priority determination mechanism: Employing a dual-mode approach of "AI automatic identification + manual review," priority is determined based on the target's threat level or importance.

[0053] Prioritization is based on importance: high-priority targets are personnel, critical equipment, or carriers of hazardous materials; medium-priority targets are ordinary mobile targets; low-priority targets are non-threatening static targets or targets in non-core areas; when the AI ​​automatically determines that the confidence level does not reach the preset threshold, manual review is triggered to ensure the accuracy of priority determination.

[0054] ② Preset multiple scheduling and sub-machine type proportion control

[0055] Preset multiple dynamic adjustment: Based on the mission scenario and target density, the preset multiple of the sub-machine scheduling is dynamically adjusted to ensure that the number of sub-machines can cover the target requirements, while avoiding sub-machine redundancy; the total number of sub-machines is the result of the target number and the preset multiple rounded up, and does not exceed the sum of the maximum load capacity of the mother machine and the maximum capacity of the launcher;

[0056] Sub-machine type proportion control: High-speed assault-type racing drones should account for no less than 40%, and medium- and low-speed precision AI racing drones should account for the remaining proportion.

[0057] The ratio should not exceed 60%, ensuring that both high-priority and medium-to-low-priority targets have suitable sub-machines for coverage, avoiding insufficient or redundant sub-machines of a single type.

[0058] ③ Type adaptation allocation rules

[0059] Based on the target identification results, and combined with core factors such as target priority, motion characteristics, target and sub-machine position, and urgency (threat level), differentiated sub-machine allocation rules are formulated and conflict handling mechanisms are clarified:

[0060] High-priority targets: Based on the location matching degree between the target and the sub-machine, and the urgency of the target (threat level), high-speed assault drones are prioritized for allocation to ensure rapid response and interception; if the location matching degree of high-speed assault drones is low or there are no available resources, the allocation strategy will be adjusted after comprehensive evaluation.

[0061] Medium-priority targets: Based on the target's motion state (high-speed / low-speed movement or static), and combined with the target's position and the urgency of the sub-machine, high-speed moving targets are preferentially matched with high-speed sub-machines in suitable positions, while low-speed moving or static targets are preferentially matched with AI sub-machines.

[0062] Low-priority objectives: Allocate idle submachines on demand, prioritize the use of AI submachine resources with matching locations, and avoid occupying high-speed submachines; only consider calling high-speed submachines when no AI submachines are available and the location of high-speed submachines is suitable.

[0063] Conflict handling: When resources for a certain type of submachine are insufficient, a type substitution mechanism is triggered. This mechanism combines the target location and urgency to select other types of submachines with suitable performance for allocation, ensuring that no target is missed.

[0064] ④ Launch / Drop Mode Switching Control

[0065] Mode determination: Switch deployment mode according to target distribution: Single or dispersed targets use "single precision drop / launch" mode to ensure that the sub-units arrive at the target area accurately; clustered targets use "multiple batch drop / launch" mode to shorten the time difference for sub-units to arrive at the target area;

[0066] Mode execution: When the mother machine drops AI slave units, in single-machine mode, they are unlocked and dropped sequentially, while in batch mode, they are unlocked synchronously to avoid mid-air collisions. When the launcher launches high-speed slave units, in single-machine mode, they are launched independently, while in batch mode, they are launched synchronously to ensure launch efficiency and safety.

[0067] (3) Dual-link collaborative control mechanism: stable interaction and emergency support

[0068] The system adopts a dual-backup design of "MESH networking wireless link + magnetic interface wired link", and ensures communication stability through timing coordination and intelligent switching. The specific design is as follows:

[0069] ① Dual-link function positioning

[0070] MESH wireless link: operates in an anti-interference frequency band, supports long-distance data transmission, is used for critical status data backhaul and core command issuance, and has the ability to adaptively adjust the channel to reduce the impact of environmental interference;

[0071] Magnetic interface wired link: It has high-speed data transmission capability and is used for target coordinate pre-transmission, mission command loading and start signal triggering before throwing / launching. The transmission is stable and not easily affected by interference.

[0072] ② Dual-link interactive timing coordination

[0073] Pre-transmission phase: When the slave unit is still attached to the carrier, the core data is pre-transmitted and the start signal is triggered through the magnetic interface to ensure that the slave unit obtains key mission information;

[0074] Execution phase: After the sub-unit detaches from the carrier, it automatically switches to the MESH wireless link to transmit flight status and target lock status in real time, ensuring that the ground terminal can keep track of the mission progress in real time;

[0075] Emergency switchover: When the MESH link transmission quality fails to meet the preset requirements, the slave unit executes the last received instruction and triggers a link abnormality alarm. The ground terminal then dispatches a backup slave unit to take over.

[0076] (4) Closed-loop scheduling throughout the entire process: complete coverage from detection to interception.

[0077] A closed-loop process of "target detection - command issuance - slave execution - status feedback - missed interception" is constructed to ensure that no targets are missed. The specific steps are as follows:

[0078] ① Target detection and data upload

[0079] Detection process: The mothership flies to the target area and scans and covers it according to the preset pattern. The AI ​​recognition module preprocesses the collected images, extracts features and classifies the targets, and calculates the target coordinates.

[0080] Data Upload: The host unit uploads the collected target information to the ground terminal via the MESH link to ensure that the ground terminal has real-time knowledge of the target distribution.

[0081] ② Command issuance and carrier response

[0082] Command generation: After receiving target information, the ground terminal generates scheduling commands according to resource allocation logic, specifying key information such as sub-unit type, launch / drop mode, and target parameters;

[0083] Command transmission: Commands generated by the control software are converted via terminal protocol and then forwarded to the host unit or transmitter.

[0084] Carrier response: After receiving the command, the mother unit adjusts to a stable attitude and triggers the release unit to unlock; after receiving the command, the launcher starts the power supply and opening mechanism to prepare for launch.

[0085] ③ Submachine execution and target interception

[0086] AI-powered racing drone execution: After being launched, it quickly achieves attitude stabilization, flies according to pre-transmitted coordinates, and activates AI for precise identification when it approaches the target at a preset distance. After locking onto the target, it adjusts its trajectory to complete the interception.

[0087] High-speed racing drone execution: After launch, the dual-mode guidance module is activated, and the drone flies according to the pre-transmitted coordinates. When it approaches the target at a preset distance, it fuses multi-modal data, locks onto the target, and quickly adjusts its attitude to complete the interception.

[0088] ④ Status feedback and real-time monitoring

[0089] Sub-unit data transmission: Real-time transmission of flight status, target lock status, mission progress, and other data to ground terminal control software.

[0090] Status indicators are updated in real time;

[0091] Ground monitoring: Operators can view the trajectory of the sub-machine and the target recognition results through control software, and can manually intervene to adjust the sub-machine task parameters to ensure the accuracy of task execution.

[0092] ⑤ Mechanism for intercepting and blocking targets that slip through the net

[0093] The following conditions will trigger a re-interception: The target has not been tracked by any slave unit for a period of time exceeding the preset duration; the slave unit sends back a "mission failed" signal; the ground terminal detects a newly added unassigned target;

[0094] Interception scheduling logic: Interception slaves are scheduled according to the priority of "idle AI slave > AI slave that has been fully charged > idle high-speed slave > high-speed slave that has been fully charged". Interception commands are issued quickly, and the interception slave flies to the target area along the shortest path to complete the interception.

[0095] Results feedback: After the interceptor completes its task, it sends back the results. The unsuccessful interceptors can be replaced by the next lower priority interceptors.

[0096] 2. Mother-Daughter UAV Cluster Scheduling System

[0097] This system serves as the hardware platform for the aforementioned method. All components collaborate through standardized interfaces and protocols. The functional configurations of the core components are as follows:

[0098] (1) Heavy-duty machine tool

[0099] Core functions: It has preset payload and endurance capabilities and can resist wind interference of preset levels; it integrates a flight control system, MESH relay module and multi-channel drop unit, each of which is equipped with a magnetic interface to support the mounting, data pre-transmission and precise drop of AI racing drones; it is equipped with a target detection module and AI recognition unit, which can complete the initial target identification and positioning.

[0100] Key features: The fuselage is made of high-strength, lightweight materials, and the drop unit is equipped with a buffer mechanism to prevent collisions when the aircraft detaches, ensuring stable initial flight attitude; the MESH relay module supports primary and backup link backup and also has relay forwarding functions to ensure uninterrupted data transmission.

[0101] (2) Medium- and low-speed precision AI racing drone

[0102] Core functions: It has target recognition and guidance capabilities, flight speed adapted to the needs of precision operations, and battery life to meet mission execution requirements; it is equipped with an image acquisition module and AI recognition algorithm, supports the recognition of two types of targets, "people" and "vehicles", and has the ability to suppress dynamic blur and re-recognize targets that are occluded;

[0103] Key features: The fuselage is equipped with a standardized magnetic interface, which is physically compatible with the magnetic interface of the mother machine's throwing unit, supporting rapid docking and data transmission; it adopts a lightweight design, which is suitable for mother machine mounting and aerial throwing.

[0104] (3) High-speed assault racing machine

[0105] Core Functions: Featuring a streamlined structural design, it possesses high-speed flight and rapid assault capabilities, with endurance sufficient for short-range, high-speed operations; equipped with a multi-modal guidance module, it adapts to the target recognition requirements of high-speed flight scenarios, enabling it to approach targets...

[0106] Initiate terminal guidance within a preset distance;

[0107] Key features: The tail section has a standardized magnetic interface that is compatible with the slot interface of the launcher; it also has a reserved expansion mounting position for the addition of a proximity fuse and warhead to improve interception effectiveness.

[0108] (4) Launch pad

[0109] Core functions: It has multiple independent transmission slots, each slot integrating a magnetic interface, a power supply interface and a guiding mechanism, supporting the storage, automatic charging and discharging and vertical transmission of high-speed slave units; it integrates a power management module with overcharge and over-discharge protection functions; it integrates an environmental monitoring unit and a safety protection module, which can monitor environmental parameters and trigger abnormal alarms.

[0110] Key features: The outer shell is made of fireproof and waterproof materials, providing a good level of protection; the cover opens automatically during launch and closes automatically after launch; the guiding mechanism ensures the accuracy of the launch direction and avoids initial flight deviation of the sub-unit.

[0111] (5) Ground control terminal

[0112] Operating system terminal: It has protocol conversion, data forwarding and command processing functions, and supports communication adaptation with interactive terminals, master units and slave units; it connects to the transmitter via wired means to ensure stable transmission of commands and data;

[0113] Domestically produced interactive terminal: supports wired and wireless dual-mode connection, with fast wireless pairing speed, suitable for outdoor operation scenarios; clear screen display, supports touch operation, and facilitates real-time monitoring and command issuance by operators;

[0114] Dedicated control software: Developed based on a domestically produced operating system, it has functions such as device management, target operation, command issuance and data feedback. It generates commands quickly and supports task log storage and review data export.

[0115] (6) Collaboration Module

[0116] MESH communication module: Supports long-distance anti-interference communication, supports multiple nodes to access simultaneously, and ensures interconnection between the master unit, slave unit, and ground terminal;

[0117] Resource allocation algorithm module: Based on target requirements and submachine performance, quickly calculate the optimal allocation scheme to ensure the rational utilization of submachine resources;

[0118] Dual-link switching module: Real-time monitoring of link transmission quality, triggering emergency switching to ensure communication stability;

[0119] The interception algorithm module quickly identifies targets that have slipped through the net and schedules interception sub-machines according to priority to ensure that no target is missed.

[0120] (III) Beneficial Effects

[0121] This invention, through an integrated design of a three-layer collaborative architecture, dynamic resource allocation, dual-link collaboration, and closed-loop scheduling, achieves significant improvements in task efficiency, resource utilization, stability, and adaptability compared to existing technologies, as detailed below:

[0122] 1. Task coverage efficiency is significantly improved, and multi-target handling capabilities are enhanced.

[0123] Expanded coverage: By adopting a hierarchical architecture of "airborne mothership + two types of daughterships", the limitations of the traditional two-level architecture are broken, the mission coverage of a single system is significantly expanded, and it can deal with more dispersed targets and clustered targets at the same time, greatly improving the multi-target handling capability.

[0124] Highly efficient cluster target handling: The multi-batch launch / drop mode shortens the time difference between the sub-units and the cluster target area, avoids target dispersion, improves interception efficiency, and solves the inefficiency problem of traditional single-unit sequential deployment.

[0125] 2. Optimized resource utilization, fully realizing the value of high-performance slave units.

[0126] Significantly reduced waste rate: Based on the precise matching logic of "target-submachine", the resource mismatch of using high-performance submachines for low-value targets and low-performance submachines for high-demand targets is avoided, the effective utilization rate of submachines is significantly improved, and the overall resource waste rate is significantly reduced.

[0127] Energy consumption optimization: Precise allocation avoids unnecessary flight of the sub-machine and power redundancy, increases the number of missions to be executed under the same power, and makes the overall energy consumption of the system more reasonable.

[0128] 3. Improved communication stability and significantly reduced task response latency.

[0129] Enhanced anti-interference capability: The dual-link backup mechanism significantly reduces the risk of communication interruption. In complex environments such as electromagnetic interference and terrain obstruction, the packet loss rate is significantly reduced, ensuring stable transmission of commands and data.

[0130] Reduced response latency: Core instructions and non-critical data are transmitted in layers to avoid bandwidth congestion. The delay in instruction issuance and interception response is significantly reduced, meeting real-time scheduling requirements and preventing target escape due to latency.

[0131] 4. The task loop is complete, and the missed interception rate is significantly reduced.

[0132] Effective solution to the problem of missed targets: The closed-loop process and the supplementary interception mechanism ensure that missed targets can be dealt with in a timely manner. Whether it is an untracked target, a target whose sub-machine task has failed, or a newly added target, the supplementary interception mechanism can respond quickly, and the missed target rate is greatly reduced.

[0133] Improved fault tolerance: The backup mechanism after a slave task fails reduces the impact of a single slave failure on the overall task, significantly improving the overall fault tolerance of the system and enhancing the reliability of task execution.

[0134] 5. Highly adaptable to various scenarios, universally applicable without customization or modification.

[0135] Multi-scenario compatibility: By dynamically adjusting the preset multiplier and launch mode, it can adapt to diverse scenarios such as defense interception, security control, and emergency detection without modifying the hardware structure, and does not require customized modifications for a single scenario;

[0136] Excellent environmental adaptability: Each component of the system has wide temperature range, waterproof, wind resistance and anti-interference capabilities, and can work stably in complex terrain and harsh environments, adapting to various operating scenarios such as mountains, water, and cities.

[0137] 6. Improved ease of operation and reduced reliance on manual labor.

[0138] High degree of automation: AI automatically identifies target priority, automatically allocates sub-machines, and automatically triggers interception, reducing the number of manual interventions and the workload of operators. Even non-professional operators can quickly get started.

[0139] Improved review efficiency: Dedicated control software automatically stores task logs and review data, supports exporting visual reports, shortens task review time, facilitates rapid optimization of task strategies, and improves the efficiency of subsequent task execution. Attached Figure Description

[0140] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0141] Figure 1 is a flowchart of the mother-daughter UAV swarm scheduling method and system provided in the embodiments of this application.

[0142] Figure 2 is a diagram showing the overall architecture of the parent-child UAV swarm scheduling method and system provided in this application embodiment. Detailed Implementation

[0143] This embodiment, in conjunction with Figure 1 (process steps 101-106) and technical principles and algorithms, provides a detailed description of the implementation process of the present invention, elaborates on the entire implementation process of the present invention, focuses on the core mechanism of hierarchical collaborative scheduling, data interaction logic and task closed-loop implementation, and highlights the innovation and feasibility of the technical solution.

[0144] I. Deployment Phase

[0145] The deployment phase is the core pre-task of the mother-daughter UAV cluster scheduling method and system in this implementation method. Its core objective is to enable the mother UAV, daughter UAVs, and ground terminals to form a collaborative organic whole through systematic state initialization, hardware function verification, and communication link construction, ensuring that each component is in a stable, reliable, and collaborative working state before the task is executed.

[0146] The core logic of this phase follows three principles: "layered verification, closed-loop feedback, and collaborative readiness." "Layered verification" refers to completing hardware self-tests and functional calibrations sequentially according to the hierarchy of "core components → subsystems → overall collaboration." "Closed-loop feedback" means that each verification step includes a closed loop of "execution → detection → judgment → adjustment" to ensure that problems are discovered and handled in a timely manner. "Collaborative readiness" refers to the final establishment of a communication link to achieve state synchronization and command interaction capabilities among various components, laying the foundation for subsequent tasks such as target detection and submachine scheduling.

[0147] The following section elaborates on the principles and process logic of the deployment phase, taking into account the functional characteristics of each component:

[0148] 1. The mother machine is ready for operation and the target scanning is complete.

[0149] As an "airborne mothership," the mothership undertakes core functions such as aircraft loading, initial target detection, and relay communication. The core of its status preparation is to achieve "stable loading, accurate navigation, and detection readiness." The process logic revolves around three core stages: hardware self-check, takeoff cruise, and AI recognition module initialization. Each stage is progressive and mutually supportive.

[0150] The principle and process logic of the mother machine hardware self-test

[0151] Hardware self-testing is a prerequisite for the safe takeoff and stable operation of the mothership. Its core principle is to eliminate initial component errors, verify the integrity of hardware functions, and ensure that core components such as sensors, power systems, and mounting units are free from potential faults through "multi-dimensional verification, closed-loop algorithm calibration, and precise fault location." The process logic is as follows:

[0152] The principle and logic closed loop of sensor self-test

[0153] Sensors are the core of the machine tool's ability to perceive its own attitude, position, and environment. The core logic of self-testing is "error source identification → algorithm calibration → accuracy verification". Targeted calibration eliminates systematic errors and ensures the accuracy of measurement data.

[0154] Error source identification logic: The measurement errors of the sensor mainly originate from zero bias error (output offset when static), environmental interference error (magnetic field, temperature interference), and installation error (coordinate axis offset). In the self-test phase, the error type is first identified through a combination of "static placement + dynamic testing"—the data collected during static placement is used to determine the zero bias error, and the dynamic test (circular motion of the magnetometer) is used to determine the environmental interference error and installation error.

[0155] The core logic of the calibration algorithm:

[0156] IMU calibration: The elimination of zero bias error is based on the "statistical averaging principle". By collecting sufficient static data, the mean filtering algorithm is used to cancel random noise, obtain the zero bias estimate and make compensation. The installation error is calculated by "attitude calculation". Based on the known motion attitude (horizontal static position), the deviation between the sensor coordinate axis and the body coordinate axis is calculated and the rotation matrix is ​​compensated.

[0157] Magnetometer calibration: Hard iron interference (superimposed from a fixed environmental magnetic field) manifests as a shift in the magnetic field origin, while soft iron interference (magnetic field distortion caused by metallic structures) manifests as a non-uniformity in the magnetic field distribution. The calibration logic involves collecting magnetic field data through omnidirectional motion, fitting a magnetic field distribution model, calculating interference compensation parameters, and restoring the distorted magnetic field distribution to an ideal spherical distribution, thus eliminating the influence of interference.

[0158] Barometer calibration: Temperature changes can cause the pressure-altitude mapping relationship to shift. The calibration logic is to establish a three-dimensional mapping model of "pressure-temperature-altitude", and by collecting pressure data at different temperatures, fit the mapping curve to achieve compensation for the measurement error caused by temperature.

[0159] Accuracy verification logic: After calibration, the accuracy is verified through "consistency check" - the calibrated measurement data is compared with the reference standard (GPS positioning altitude data, theoretical data of known attitude). If the deviation is within the preset reasonable range, the calibration is deemed valid; if the deviation exceeds the range, the calibration process is re-executed, forming a closed loop of "calibration → verification → recalibration" until the accuracy requirements are met.

[0160] The principle and logical closed loop of power system self-test

[0161] The core objective of the power system self-test is to verify the collaborative working capability of the motor and ESC, and to ensure the consistency between the motor output and the control commands. Its process logic revolves around "command issuance → response acquisition → matching verification → anomaly handling".

[0162] Command issuance logic: The flight control system issues control commands according to the strategy of "sequential start + gradient command". Sequential start is to avoid power supply fluctuations caused by the simultaneous start of multiple motors, and gradient commands are to cover the low speed operating range of the motors and ensure stability verification during the start-up phase.

[0163] Response acquisition and matching verification logic: After receiving the command, the ESC drives the motor to rotate, and the flight control system receives feedback from the ESC.

[0164] The signal acquisition measures the actual speed of the motor. The core of the matching verification is the "command-response linearity test"—determining whether the motor speed changes linearly with the control command. If the linearity meets the requirements and the deviation between the actual speed and the theoretical speed is within a reasonable range, the matching is considered normal. If there is no speed response, excessive deviation, or nonlinear change, a "power abnormality" alarm is triggered.

[0165] Anomaly Handling Logic: After an alarm is triggered, the system executes the "fault isolation → secondary verification" process. First, the faulty motor is identified by starting the faulty motor separately to eliminate interference from multiple motor linkages. Then, the power supply line and ESC connection status are checked. If it is a connection problem, the system attempts to automatically recover (reinitialize the ESC). If it is a hardware failure, the system locks the faulty motor and prompts the operator to repair it to ensure takeoff safety.

[0166] The principle and logic loop of the mount unit self-test

[0167] The core of the mounting unit self-test is to verify the "physical fixation reliability" and "data interaction continuity" to prevent the slave unit from falling off or the data transmission from being interrupted. The process logic adopts a "dual verification + redundancy guarantee" design.

[0168] The electromagnetic latch's on / off verification logic: The physical fixation reliability of the electromagnetic latch is guaranteed by a triple logic of "energization closure → status feedback → mechanical verification". After energization closure, the latch's built-in position sensor provides feedback on the closure status, and the flight control system simultaneously verifies it through "current detection"—there is a significant difference in the operating current when the latch is closed and open. By checking the current change, the authenticity of the closure status is confirmed, avoiding misjudgments caused by position sensor failure.

[0169] The data link verification logic of the magnetic interface is as follows: The continuity of the data link is achieved through a closed loop of "test data packet sending-echoing-verification". The flight control system generates a test data packet containing a specific identifier, random data, and a checksum, and sends it to the slave unit through the magnetic interface. After receiving it, the slave unit sends it back as is (echoing). The flight control system compares the consistency between the sent data packet and the echoed data packet (verified by the checksum). If they match, the link is considered to be working; if they do not match, the test data packet is sent repeatedly (up to a preset limit). If they still do not match, the link is considered to be abnormal, and an alarm is triggered.

[0170] The self-test result summary logic is as follows: After each component completes its self-test, the flight control system generates a "self-test result matrix," which includes the "normal / abnormal" status of each component and details of any abnormalities. If all components are normal, a "self-test passed" signal is sent back to the ground terminal, and the aircraft enters a takeoff-ready state. If any abnormalities are found, the abnormal component identifier and handling suggestions are sent back, awaiting operator intervention.

[0171] The principles and procedural logic of mothership takeoff and cruise

[0172] The core of the mothership's takeoff and cruise phases is to achieve "stable attitude control + precise path navigation". The process logic revolves around two stages: vertical climb and cruise. Each stage includes a closed loop of "control algorithm → status monitoring → dynamic adjustment".

[0173] Attitude control principles and logic for vertical climbing

[0174] The core challenge of vertical climbing is maintaining attitude stability while carrying multiple sub-units, and avoiding tilting that could damage the sub-units.

[0175] The detachment mechanism is based on a "cascaded PID control + anti-interference compensation" design.

[0176] The core logic of cascade PID control is as follows: it adopts a cascade structure of "outer loop height control + inner loop attitude control". The outer loop uses the climbing height as the control target and calculates the required attitude adjustment commands; the inner loop uses the attitude angle as the control target and converts the attitude adjustment commands into motor control signals. The advantage of this structure is that it decouples height control from attitude control, reduces the control complexity of a single loop, and improves stability—when the height deviation is large, the outer loop prioritizes adjusting the attitude reference value, and the inner loop quickly responds to the attitude deviation, ensuring a smooth climbing process.

[0177] Anti-interference compensation logic: During vertical ascent, external interference such as wind may cause attitude fluctuations. The core of anti-interference compensation is "interference observation + reverse cancellation"—the attitude change rate is collected in real time by the IMU to determine the direction and intensity of the interference, and a reverse control signal is generated to cancel the impact of the interference on the attitude. For example, if the fuselage is detected to be tilted to one side, the output power of the motor on the corresponding side is immediately increased to correct the tilted attitude.

[0178] Status monitoring and adjustment logic: During the climb, the flight control system collects status data such as altitude, attitude angle, and motor output power in real time. If the altitude deviation exceeds the preset range, the outer loop PID parameters are adjusted (increasing the proportional coefficient) to accelerate the climb speed. If the attitude fluctuation is too large, the inner loop PID parameters are adjusted (increasing the derivative coefficient) to enhance damping and suppress fluctuations, forming a dynamic optimization closed loop of "monitoring → adjustment → re-monitoring".

[0179] Navigation principles and logic of route cruise

[0180] The core objective of route navigation is to fly precisely to the target area along a preset route. Its logic revolves around "path planning → positioning fusion → speed control" to ensure the accuracy and stability of navigation.

[0181] The core logic of path planning is as follows: the preset route consists of discrete waypoints, and the purpose of path planning is to transform discrete waypoints into a smooth flight path, avoiding abrupt attitude changes between waypoints. A "continuous path interpolation algorithm" is used to construct a smooth curve based on waypoint coordinates, ensuring the continuity of the first derivative (velocity) and second derivative (acceleration) of the path, thus enabling a smooth transition in the flight attitude of the mother aircraft.

[0182] Speed ​​control logic: The selection of cruise speed needs to take into account both flight efficiency and stability. The control logic is "adaptive speed adjustment" - dynamically optimizing the speed based on the current flight status (attitude stability, navigation accuracy): if the flight is stable and the navigation accuracy is high, the speed is appropriately increased; if the flight is unstable or the navigation accuracy is low, the speed is reduced to ensure the safety of the cruise process; at the same time, the speed adjustment adopts the "ramp transition" method to avoid attitude fluctuations caused by sudden speed changes.

[0183] 3. The principle and process logic of AI recognition module initialization

[0184] The core goal of initializing the AI ​​recognition module is to enable the module to have accurate target detection capabilities. The process logic revolves around "model loading → feature library synchronization → parameter configuration" to ensure that the module's computing power is suitable, features are complete, and data acquisition is compatible.

[0185] Optimization principles and logic of model loading

[0186] The core of loading AI recognition models is "computing power adaptation + fast startup," to avoid the model's computational load exceeding the main control chip's processing power.

[0187] The loading time may be too long, affecting the progress of the task.

[0188] The core logic of model optimization is to adopt a "pruning + quantization" optimization strategy. Pruning removes redundant convolutional kernels (those that contribute little to feature extraction) to reduce computation. Quantization converts model weights from high-precision floating-point numbers to low-precision integers, reducing memory usage and computational complexity. During optimization, "precision loss verification" ensures that the optimized model's detection accuracy meets the target detection requirements, forming a closed loop of "optimization → verification → further optimization."

[0189] Model loading logic: To avoid memory overflow during loading, a "block loading + sequential initialization" strategy is adopted. The optimized model is divided into multiple modules according to the layer structure and loaded sequentially according to the inference order. After the previous module is loaded and initialized, the next module is loaded. At the same time, the temporary loaded data of the previous module is released, and only the core parameters required for inference are retained to ensure memory stability during the loading process.

[0190] The consistency principle and logic of feature database synchronization

[0191] The target feature library is the foundation of model recognition. The core logic of synchronization is "integrity guarantee + consistency verification" to ensure that the feature library acquired by the AI ​​recognition module is completely consistent with the feature library stored in the host machine.

[0192] The construction logic of the feature library: The feature library contains the core features of targets such as personnel, vehicles, and key equipment. Its construction is based on "multi-dimensional feature fusion" - extracting the contour features, texture features, and semantic features of the target, removing redundant information through dimensionality reduction algorithms, forming a compact feature vector, and storing it in binary format to improve synchronization efficiency.

[0193] The logic of the synchronization process:

[0194] Synchronization Trigger: After the model is loaded, the AI ​​recognition module sends a synchronization request to the host machine. The request includes the feature library version and category supported by the module.

[0195] Difference detection: The host machine compares the version of the local feature library with the version requested by the module. If the versions are the same and the categories are complete, no synchronization is required; if the versions are different or the categories are missing, the synchronization process is initiated.

[0196] Chunked transmission: The feature library is divided into blocks of a preset size and transmitted using an "acknowledgment and retransmission" mechanism—the sender sends one data block and waits for the receiver's acknowledgment signal before sending the next data block, thus avoiding data loss.

[0197] Consistency verification: After all data blocks have been transmitted, the AI ​​recognition module calculates the check value of the feature library and compares it with the check value of the master machine. If they match, the synchronization is considered valid; if they do not match, it requests the retransmission of missing or erroneous data blocks to ensure the integrity of the feature library.

[0198] Adaptive principle and logic of image acquisition parameter configuration

[0199] The core of configuring image acquisition parameters (exposure time, acquisition frame rate, etc.) is "adapting to ambient lighting + ensuring image quality", and its logic is based on "environmental perception → parameter matching → effect verification".

[0200] Ambient lighting perception logic: By acquiring global brightness information from the initial image, it determines the ambient lighting level (low light, high light, low light).

[0201] (Normal light, strong light) - Calculate the average gray value of the image. If the average gray value is lower than the preset threshold, it is determined to be a weak light environment. If it is higher than the preset threshold, it is determined to be a strong light environment. If it is in between, it is a normal light environment.

[0202] Parameter matching logic:

[0203] Exposure time: In low light conditions, the exposure time is extended to improve image brightness; in strong light conditions, the exposure time is shortened to avoid overexposure; at the same time, the adjustment of the exposure time is subject to "dynamic blur constraint"—based on the flight speed of the mothership, the maximum allowable exposure time is set to ensure that the image of the moving target is not significantly blurred.

[0204] Acquisition frame rate and resolution: Reduce acquisition frame rate and resolution in low light environment to reduce noise impact; increase acquisition frame rate and resolution in normal light environment to ensure target detail capture; the matching of acquisition frame rate and resolution follows the "computing power adaptation principle" to ensure that the computational load of image acquisition and subsequent processing does not exceed the computing power limit of the main control chip.

[0205] Effect verification logic: After the parameters are configured, a frame of image is acquired for effect verification - to determine whether the brightness of the image is appropriate, whether the target outline is clear, and whether the noise is within a reasonable range. If the effect meets the requirements, the configuration is deemed effective; if the effect is poor (overexposure, underexposure, excessive noise), the parameters are readjusted to form a closed loop of "configuration → verification → reconfiguration" to ensure that the image quality meets the target detection requirements.

[0206] 2. Sub-unit preparation principle and process logic of the launcher

[0207] As an integrated "storage, transportation, and launch" carrier for high-speed racing drones, the core objective of the launch pad during the preparation phase is to achieve "sufficient energy for the drones, data readiness, and environmental safety." The process logic revolves around two core aspects: high-speed racing drone charging management and environmental monitoring and safety protection. Each aspect reflects the design concept of "safety first, closed-loop control."

[0208] The principles and process logic of charging management for high-speed racing drones

[0209] The core of charging management is "safe charging + sufficient energy" to avoid safety risks caused by overcharging, over-discharging, and overheating of batteries. The process logic is based on "charging stage division + anomaly protection + status feedback".

[0210] The principle and logic of dividing charging stages

[0211] The charging process is divided into a constant current charging stage and a constant voltage charging stage. The core basis for this division is the battery's charging characteristic curve. When the battery voltage is lower than a preset threshold, constant current charging is used to quickly replenish the battery energy. When the battery voltage is close to the full charge voltage, constant voltage charging is switched to avoid overcharging due to excessive voltage.

[0212] The control logic during the constant current charging phase is as follows: Current feedback control maintains a stable charging current, ensuring rapid battery energy replenishment while preventing battery overheating caused by current fluctuations. The current control employs a "PID control algorithm," which collects the charging current in real time, compares it with a preset constant current value, and adjusts the output voltage of the charging circuit to offset current fluctuations caused by changes in line resistance and battery internal resistance.

[0213] Control logic during constant voltage charging: When the battery voltage reaches a preset switching threshold, the charging mode automatically switches to constant voltage charging to maintain a stable charging voltage. At this time, the battery charging current gradually decreases, and when the current decreases to a preset cutoff value...

[0214] When the threshold is reached, the battery is determined to be fully charged, charging is stopped, and the system switches to float charging mode (to maintain battery voltage and prevent voltage drop). This constitutes the closed-loop logic for charging protection.

[0215] The core of charging protection is "risk prediction + rapid response". By monitoring the battery status and environmental status in real time, abnormal situations can be handled in a timely manner to avoid safety risks.

[0216] Temperature protection logic: Battery charging efficiency and safety performance are significantly affected by temperature. Excessive temperature can lead to thermal runaway, while excessively low temperature can reduce charging efficiency and damage the battery. The protection logic monitors the battery temperature in real time. If the temperature exceeds the preset safe range, charging is immediately paused; once the temperature returns to the safe range, charging is restarted, forming a closed loop of "monitoring → pausing → resuming".

[0217] Overcharge and over-discharge protection logic: Overcharge protection employs a dual system of "voltage monitoring + time redundancy"—charging stops when the battery voltage reaches the full charge voltage and the current drops to the cutoff threshold; if voltage monitoring fails, charging time redundancy is used for determination (a maximum charging time is set, and charging is forcibly stopped if this time is exceeded). Over-discharge protection detects the battery voltage before charging; if the voltage is below the over-discharge threshold, the battery is deemed damaged, triggering a "battery abnormality" alarm, prohibiting charging and transmission, and avoiding safety risks caused by over-discharged battery charging.

[0218] Status feedback logic: During the charging process, the power management module transmits the charging status (charging stage, current voltage, current, temperature) back to the ground terminal in real time, which is convenient for the operator to monitor in real time; after charging is completed, it transmits the "sub-unit charging completed" signal, and the high-speed racing drone enters the standby state, waiting for subsequent data preloading and transmission commands.

[0219] Principles and Process Logic of Environmental Monitoring and Safety Protection

[0220] The core objective of environmental monitoring and safety protection is to ensure the storage safety of the launch pad and sub-units, and to avoid equipment damage or safety accidents caused by environmental anomalies. The process logic revolves around "real-time monitoring → risk assessment → graded response".

[0221] The principles and logic of environmental monitoring

[0222] The core of environmental monitoring is "multi-dimensional perception + data credibility verification" to ensure the accuracy of monitoring data and provide a reliable basis for safety protection.

[0223] Temperature and humidity monitoring logic: Temperature and humidity sensors collect real-time temperature and humidity data inside the launch pad. The collection cycle is set based on "risk response speed" and "energy consumption balance"—ensuring that abnormal situations can be detected in a timely manner while avoiding energy waste caused by frequent collection. The collected data is processed through "moving average filtering" to cancel random noise and improve data reliability.

[0224] Smoke / Particulate Matter Monitoring Logic: Smoke detectors determine the presence of fire risk by detecting the concentration of particulate matter in the air. This is based on the light scattering effect—particulate matter scatters light emitted by the sensor, and the intensity of the scattered light is positively correlated with the particulate matter concentration. The monitoring logic involves real-time acquisition of the scattered light intensity and comparison with a preset alarm threshold to determine the presence of a safety risk.

[0225] Hierarchical response logic for security protection

[0226] The security protection adopts a "tiered response" strategy, taking different protective measures according to the risk level to ensure that the risk is effectively controlled while avoiding mission interruption caused by over-protection.

[0227] Warning-level response: When temperature and humidity data exceed the normal range but do not reach the danger threshold, a warning-level response is triggered—an "environmental anomaly" warning message is sent to the ground terminal to remind the operator to pay attention to environmental changes; at the same time, the ventilation status of the transmitter is adjusted (ventilation vents are opened) to try to improve the internal environment.

[0228] Emergency Response: When the temperature and humidity data reach the danger threshold, or the smoke / particulate matter concentration exceeds the alarm threshold, an emergency response is triggered—the power supply circuit of the launcher is immediately cut off (to avoid exacerbating the risk of short circuits in electrical equipment), and an audible and visual alarm is triggered (to prompt the operator to take emergency action); if the smoke / particulate matter concentration continues to rise, the "emergency pressure relief" mechanism is further triggered (the launcher hatch is opened to release smoke and heat), minimizing safety risks.

[0229] Recovery logic: Once the environmental monitoring data returns to a safe range, the system executes the "safety verification → step-by-step recovery" process—first verifying the insulation of the power supply circuit and the integrity of the equipment status. If there are no abnormalities, the power supply circuit is reconnected and the alarm is canceled; if there are equipment abnormalities, the abnormal area is locked and the operator is prompted to perform maintenance.

[0230] 3. Communication connection principle and process logic between ground terminal and system

[0231] As the core of mission control, the ground terminal's core communication connection goal is "stable interconnection + status synchronization" to ensure smooth command interaction and data transmission between the ground terminal and the host and launch pad. The process logic revolves around two core links: MESH network link establishment and ground terminal APP status synchronization.

[0232] The principles and process logic of MESH network link establishment

[0233] The core advantage of MESH networking is "self-organization and self-repair". Its link establishment principle revolves around "node discovery → link selection → link maintenance", ensuring the stability and reliability of communication links.

[0234] Interaction logic for node discovery

[0235] Node discovery is a prerequisite for networking. The core is to enable mutual identification and information exchange between devices. The process logic is based on "beacon broadcasting → response feedback → information synchronization".

[0236] Beacon broadcast logic: The MESH repeaters of the ground terminal send beacon frames at fixed intervals. The beacon frames contain core information such as the repeater's device identifier, operating frequency band, signal strength, and load status. The broadcast period is set based on a balance between "discovery speed" and "channel occupancy"—a period that is too short will lead to channel congestion, while a period that is too long will prolong the node discovery time.

[0237] Response feedback logic: After the MESH module of the host and transmitter receives the beacon frame, it parses the information in the frame. If it is determined to be the same network system (verified by device identifier), it sends back a response frame. The response frame contains information such as its own device type, mounting status (number of slave devices mounted on the host), and communication capabilities.

[0238] Information synchronization logic: After receiving a response frame, the repeater stores the information of each node, establishes a "node information table", and packages...

[0239] It includes node identifiers, device types, signal strength, load status, etc., providing a basis for subsequent link selection.

[0240] Link selection weight logic

[0241] The core of link selection is "optimal link matching," which is based on a comprehensive judgment of multiple dimensions and weights to ensure that the selected link has the advantages of stable signal, low load, and low transmission delay.

[0242] Weighting logic: The weighting factors for link selection include signal strength, load status, and transmission delay, with signal strength having the highest weight (ensuring link stability), followed by load status (avoiding link congestion), and then transmission delay (ensuring real-time command interaction). The specific values ​​of each weighting factor are optimized based on the actual communication scenario, forming a dynamic weighting model.

[0243] Optimal link determination logic: Based on the data in the node information table, the repeater calculates the comprehensive weight score of each node, selects the node with the highest score as the primary communication node (master node), and the other nodes as backup communication nodes. The primary communication node is responsible for the main instruction and data transmission, and the backup communication nodes automatically switch over in the event of a primary link failure to ensure communication continuity.

[0244] Heartbeat mechanism for link maintenance

[0245] After the link is established, the link status is maintained through a "heartbeat interaction" mechanism to ensure that link anomalies are detected and dealt with in a timely manner.

[0246] Heartbeat interaction logic: The repeater and the primary communication node send heartbeat packets at fixed intervals. The heartbeat packet contains a link status identifier and data transmission confirmation information. If the repeater does not receive a heartbeat packet from the primary communication node within a preset time, it determines that the primary link is abnormal, automatically switches to the backup communication node, and re-establishes the link.

[0247] Link repair logic: After the main link fails, the repeater sends a link repair request to the original main communication node. If a response is received, it attempts to restore the main link; if no response is received, it maintains the backup link communication and sends a "link switch" prompt to the ground terminal to ensure that the operator is aware of the change in communication status.

[0248] The principle and process logic of ground terminal APP status synchronization

[0249] The core objective of status synchronization is to enable operators to monitor the working status of each component of the system in real time, providing a basis for task decision-making. The process logic revolves around "device list construction → status data transmission → image synchronization decoding".

[0250] Logic for building the device list

[0251] The device list is the foundation for operators to intuitively view system components, and its construction logic is based on "information summary → classification and sorting → status identification".

[0252] Information aggregation logic: The APP receives device information from each node in the MESH link and aggregates it to form a "device information pool", which includes information such as device identifier, device type, communication status, and working status.

[0253] Category and sorting logic: The APP categorizes devices by device type (master unit, transmitter, slave unit), and sorts them by communication status (normal / abnormal) within the same type. Devices with normal communication are displayed first, making it easier for operators to quickly focus on core devices.

[0254] Status identification logic: The APP adds a status identifier to each device. The status identifier is updated in real time based on the device's working status data, ensuring that the operator can intuitively grasp the device status.

[0255] Periodic logic of state data transmission

[0256] The transmission cycle of status data is set based on "data importance + real-time requirements" to ensure that core data is updated in real time, while non-core data balances transmission efficiency and channel occupancy.

[0257] Mother machine status transmission logic: As the core node, the mother machine's status data such as position, speed, attitude, and power are crucial for mission decisions. The transmission cycle is set to a short cycle (to ensure real-time performance) to facilitate operators' real-time monitoring of the mother machine's flight status.

[0258] Transmitter status transmission logic: The real-time requirements for data such as the slot status and environmental parameters of the transmitter are lower than those of the host machine. The transmission cycle is set to a long cycle to balance transmission efficiency and channel occupancy. However, when environmental parameters are abnormal, the transmission cycle is automatically shortened to ensure timely feedback of abnormal information.

[0259] Data parsing and display logic: After receiving status data, the APP parses it according to the preset data format and converts the raw data into intuitive visual information (attitude angle is displayed in the form of an instrument panel, and battery level is displayed in the form of a progress bar), which is convenient for operators to understand quickly.

[0260] Image synchronization decoding and display logic

[0261] The synchronization of regional images from the mother machine is the prerequisite for target detection. Its logic revolves around "data transmission → decoding processing → display optimization" to ensure the smoothness and clarity of the images.

[0262] Image data transmission logic: Image data captured by the high-definition camera of the host machine is transmitted frame by frame through the MESH link. During the transmission process, a "frame compression + verification" strategy is adopted - frame compression reduces the amount of data and improves the transmission speed; frame verification ensures the integrity of image data and avoids image distortion caused by data loss during transmission.

[0263] Decoding processing logic: After receiving the compressed image data, the APP uses an efficient decoding algorithm to restore the image. During the decoding process, "noise suppression + brightness adjustment" optimization is performed to improve image clarity. At the same time, the "frame buffering" mechanism is used to offset the image stuttering caused by transmission delay and ensure smooth image display.

[0264] Display optimization logic: The APP adaptively adjusts the image display ratio according to the screen size of the tablet to avoid image stretching and distortion; it also supports image zoom function, which makes it easy for operators to zoom in to view the details of the target area, providing clear visual support for subsequent target labeling and task decision-making.

[0265] II. Target Detection and Allocation Phase (Figure 1, Steps 101-103)

[0266] The target detection and allocation phase is the core hub of the hierarchical collaborative scheduling of the parent-child UAV swarm, connecting the "deployment phase" and the "command execution phase." Its core objective is to achieve deep adaptation between the capabilities of the child UAVs and the mission requirements through a complete process logic of "accurately perceiving target needs → scientifically scheduling child UAV resources → optimally matching supply and demand," thus paving the way for subsequent child UAV deployments.

[0267] It provides a basis for decision-making regarding launch and target interception.

[0268] This phase follows a three-layer closed-loop logic of "data-driven - algorithm decision-making - manual verification": "data-driven" refers to building a task requirement database based on the full-dimensional information of the target (quantity, coordinates, motion state, priority) collected by the AI ​​recognition module of the main machine; "algorithm decision-making" refers to transforming the target requirements into a sub-machine scheduling scheme through resource scheduling algorithms and matching optimization algorithms; "manual verification" refers to the operator confirming the rationality of the algorithm output results to avoid algorithm deviations in extreme scenarios and ensure the reliability of the scheme.

[0269] The principles and process logic of this stage are explained in detail below with reference to steps 101-103 in Figure 1:

[0270] 1. Step 101: Principles and Process Logic of Target Information Collection

[0271] Target information acquisition is the foundation of this stage. Its core is to transform raw image data into structured target requirement information through a progressive process of "image preprocessing → target detection and feature extraction → target priority determination," ensuring the accuracy of subsequent resource scheduling and matching. Its core logic is "eliminating interference → extracting features → quantifying priorities," with each step revolving around "improving the credibility of target information."

[0272] The principles and logical closed loop of image preprocessing

[0273] The core purpose of image preprocessing is to eliminate the interference of complex outdoor environments (uneven lighting, light spot noise) on target detection, optimize image quality, and lay the foundation for subsequent target feature extraction. Its workflow logic is based on a closed-loop design of "interference type identification → targeted processing → effect verification," and the specific principle is as follows:

[0274] The core principles and logic of noise reduction processing

[0275] In outdoor environments, images are susceptible to fluctuations in natural light and sensor noise, resulting in light spots and random noise that obscure the details of the target. The core principle of noise reduction processing is "utilizing the correlation of neighboring pixels," which means smoothing local pixel fluctuations through weighted averaging while preserving the edge features of the target and avoiding excessive blurring.

[0276] Interference identification logic: First, noise types are identified through "grayscale value variance analysis"—random noise manifests as sudden changes in local pixel grayscale values ​​(large variance), while spot interference manifests as an overall higher grayscale value within a region (small variance but high mean). For both types of interference, "Gaussian filtering" is used to achieve differentiated suppression: the kernel function of Gaussian filtering has the characteristic of "high weight for center pixels and low weight for edge pixels," which can smooth random noise (suppress sudden pixel changes) and moderately weaken spot interference (reduce regional mean deviation).

[0277] Filter kernel selection logic: The size of the filter kernel is designed to balance noise reduction effect with feature preservation. If the kernel is too small, noise reduction will be incomplete (unable to cover a large area of ​​light spots), while if the kernel is too large, the target edges will be blurred (loss of detailed features). This process determines the kernel size through "multi-round testing and verification": For common outdoor light spots and noise scales, the selected kernel can cover typical interference ranges, while ensuring that the gray-level gradient of the target edge is not excessively smoothed (verified by the edge detection algorithm, the gradient value retention rate of the target edge after filtering is ≥80%).

[0278] Effect verification logic: After noise reduction is completed, the effect is verified by "signal-to-noise ratio (SNR) calculation" - comparing the signal energy (gray value of the target area) and noise energy (gray value fluctuation of the background area) of the image before and after noise reduction. If the SNR is increased to the preset threshold (ensuring that the target features can be identified), then proceed to the next step; if it is not up to standard, adjust the filter kernel parameters and reprocess to form a closed loop of "processing → verification → adjustment".

[0279] The principle and logic of brightness enhancement

[0280] In outdoor scenes, uneven lighting (shadow areas, backlighting) can cause the target area to be too dark, making features difficult to clearly display. The core principle of brightness enhancement is "grayscale distribution reconstruction," which adjusts the grayscale value distribution range of the image to increase the brightness of dark areas while avoiding overexposure in bright areas, thus expanding the dynamic range of the image.

[0281] Brightness problem diagnosis logic: First, "grayscale histogram analysis" is used to locate brightness defects—an excessively high proportion of dark areas manifests as a leftward-biased histogram peak (concentration of low grayscale values), and in backlit environments, this manifests as compression at both ends of the histogram (narrow dynamic range). To address this issue, a "histogram equalization" algorithm is used to reconstruct the grayscale distribution: its core logic is to map the cumulative grayscale distribution function (CDF) of the original image to a uniform distribution, causing the grayscale values ​​in dark areas to expand into higher grayscale ranges, while moderately compressing the grayscale values ​​in bright areas, thus achieving brightness balance across the entire image.

[0282] Dynamic range optimization logic: The key to brightness enhancement is to avoid "detail loss due to over-enhancement"—simply stretching the grayscale range may lead to simultaneous amplification of noise in dark areas and pixel saturation (overexposure) in bright areas. Therefore, based on histogram equalization, a "grayscale truncation" mechanism is introduced: upper and lower limits for grayscale values ​​are set (based on the typical grayscale range of the target area). Pixels exceeding the upper limit are set to the upper limit value (to avoid overexposure), and pixels below the lower limit are set to the lower limit value (to suppress noise), ensuring that the dynamic range of the enhanced image is both expanded and controllable.

[0283] Effect verification logic: Verification is performed through dual verification of "dynamic range quantization" and "visual sharpness evaluation" - the dynamic range (the difference between the maximum and minimum grayscale values) must be more than 1.5 times that before enhancement, and the sharpness of the target edge is statistically analyzed through the edge detection algorithm (the grayscale gradient value of the edge pixels is ≥ 90% of that before enhancement) to ensure that the brightness enhancement does not damage the target features.

[0284] The principle and logic of size scaling

[0285] The AI ​​recognition model YoloV3-slim has fixed requirements for the size of the input image. The core purpose of scaling is to adapt the preprocessed image to the model's input specifications while avoiding distortion of target features. Its core principle is "pixel interpolation reconstruction," which uses a reasonable interpolation algorithm to preserve the shape and contour features of the target during the scaling process.

[0286] Scaling requirement analysis logic: The design of the model input size is based on the "balance between computing power and detection accuracy"—too small a size will lead to the loss of target features (small targets cannot be identified), while too large a size will increase the model inference time (affecting real-time performance). In this process, the scaling size must strictly match the model input specifications, while ensuring that the targets in the original image can still retain key features (limb contours of people, wheel structures of vehicles) after scaling.

[0287] Interpolation algorithm selection logic: Commonly used interpolation algorithms include nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation. Nearest neighbor interpolation can easily lead to jagged edges on the target (feature distortion), while bicubic interpolation has too much computation (affecting real-time performance). Therefore, "bilinear interpolation" is chosen. Its principle is based on the four neighboring pixels around the target pixel, and the scaled pixel value is calculated by weighted averaging. This ensures smooth target edges (without obvious jagged edges) while controlling the amount of computation (meeting the requirements of real-time detection).

[0288] Feature fidelity verification logic: After scaling, the degree of distortion is verified by "feature matching degree calculation" - compare the key features of the target before and after scaling (the coordinates of the inflection point of the contour, the gray distribution of the texture). If the feature matching degree is ≥95% (indicating that the distortion is minimal), the scaling is deemed effective; if the matching degree is insufficient, the interpolation algorithm parameters (weighting coefficients) are adjusted and reprocessed to ensure that the target features are not lost after scaling.

[0289] Principles and Process Logic of Object Detection and Feature Extraction

[0290] Target detection and feature extraction are the core steps in transforming preprocessed images into structured target information. The core logic is "multi-scale detection coverage → motion state quantification → key feature extraction" to ensure accurate target identification and acquisition of its core attributes.

[0291] The principles and logic of multi-scale target detection

[0292] In outdoor scenarios, target sizes vary significantly (people and vehicles). Single-scale detection can easily lead to either "missed detection of large targets (insufficient coverage of small-scale detection heads)" or "false detection of small targets (insufficient resolution of large-scale detection heads)." Therefore, a multi-scale detection head design is adopted, the principle of which is as follows:

[0293] The logic behind the detection head scale division: The three detection heads of the YoloV3-slim model (corresponding to different feature map scales) follow the principle of "target size adaptation"—the large-scale detection head (corresponding to a small feature map) has a large receptive field, suitable for detecting distant, large targets (vehicles); the small-scale detection head (corresponding to a large feature map) has a small receptive field, suitable for detecting close-range, small targets (people). The scale division of the three detection heads is based on "typical target size statistics": by analyzing a large number of outdoor scene samples, the pixel size range of different targets is determined, and then the corresponding detection head scale is matched to achieve full-size target coverage.

[0294] Target bounding box prediction logic: Each detection head predicts the target bounding box through "anchor box matching"—anchor boxes are typical target shapes based on sample statistics (rectangular boxes for people, rectangular boxes for vehicles). The detection head calculates the intersection-union ratio (IOU) between the anchor boxes and the actual target shapes, filters out anchor boxes with high matching degrees, and then fine-tunes the anchor box coordinates through a regression algorithm to obtain the accurate target bounding box (x1, y1, x2, y2). Simultaneously, the detection head outputs the target's category confidence score (representing the probability that the target belongs to a certain category). The confidence score threshold is set based on a "precision-recall balance"—ensuring that low-confidence suspected targets are not misclassified, and high-confidence true targets are not missed.

[0295] Detection result fusion logic: After the three detection heads output their detection results, they need to be fused using Non-Maximum Suppression (NMS).

[0296] The algorithm eliminates duplicate detection boxes—for multiple detection boxes of the same target, it retains the box with the highest confidence and deletes other boxes with high overlap (IOU ≥ preset threshold) to avoid the target being counted repeatedly. After fusion, a "target detection list" is generated, containing the bounding box, category, and confidence score of each target, providing a foundation for subsequent motion state calculation and feature extraction.

[0297] The principle and logic of motion state calculation

[0298] The target's motion state (speed, direction) directly affects the submachine response strategy (high-speed targets need to be intercepted quickly). Its core calculation is based on the "optical flow method", which quantifies the motion state by analyzing the motion trajectory of target pixels in consecutive frame images.

[0299] The core principle of optical flow is the "correlation between pixel motion and target motion"—assuming that the pixel grayscale values ​​on the target surface remain unchanged in consecutive frames, the actual speed and direction of the target's motion are inferred by calculating the displacement of corresponding pixels in adjacent frames. In this process, the "Lucas-Kanade optical flow method" is used, whose advantage lies in improving the stability of displacement calculation (resistance to noise interference) through pixel grayscale consistency constraints within a local window.

[0300] Window size selection logic: The window size in optical flow directly affects computational accuracy and anti-interference capability—a window that is too small is easily affected by noise (large fluctuations in displacement calculation), while a window that is too large cannot capture the local motion of the target (local displacement differences caused by target rotation). This process determines the window size through "motion state stability test": for common outdoor target motion speeds (people walking, vehicles driving), the selected window can cover enough pixels (ensuring the effectiveness of grayscale consistency constraints) and reflect the local motion characteristics of the target, keeping the motion speed calculation error within an acceptable range.

[0301] Motion state quantization logic:

[0302] Displacement calculation: For the same target in 3 consecutive frames (confirmed by bounding box matching), select multiple feature points (edge ​​turning points) within each target region and calculate the displacement (Δx, Δy) of each feature point in adjacent frames.

[0303] Speed ​​calculation: Based on the image frame rate (number of frames per unit time), the pixel displacement is converted into actual physical speed—the pixel-physical scale mapping relationship of the image is calculated by the height of the host machine and the focal length of the camera (1 pixel corresponds to 0.5 meters), and then combined with the time interval (1 / frame rate) to obtain the actual speed of the target (speed = displacement / time).

[0304] Direction calculation: Determine the target's direction of motion (30° east of north) by using the angle of the displacement vector (Δx, Δy).

[0305] Motion state verification logic: The calculation results are verified through "multi-frame consistency check" - if the motion speed fluctuation of 3 consecutive frames is ≤ preset threshold (indicating that the motion state is stable), the result is deemed valid; if the fluctuation is too large, feature points are re-selected for calculation to avoid errors caused by target occlusion or sudden changes in lighting.

[0306] The principles and logic of target feature extraction

[0307] Target features are the core basis for subsequent priority determination and sub-machine matching. It is necessary to extract "key features" that can distinguish target categories (human body contours, vehicle wheel textures), which is based on the principle of "combining low-level visual features with high-level semantic features".

[0308] Feature type selection logic:

[0309] Contour Features: The shape and contour of the target are the most intuitive features for distinguishing categories (people are upright rectangles, vehicles are long rectangles). The edge pixels of the target are extracted by the "edge detection algorithm" (Canny algorithm), and then a smooth contour curve is obtained by "contour fitting". The key parameters of the contour (perimeter, area, aspect ratio) are stored.

[0310] Texture features: The texture differences of the target surface are an important basis for subdivision (metal texture of vehicles, fabric texture of clothing of people). The contrast, correlation and other parameters of the texture are extracted by the "Gray Co-occurrence Matrix (GLCM)" - the Gray Co-occurrence Matrix reflects the spatial distribution relationship of pixels with different gray values ​​in the image, and its statistical parameters can quantify the roughness and regularity of the texture.

[0311] Semantic features: Based on the intermediate layer output of the AI ​​model, feature vectors that can represent the semantics of the target (output of the convolutional layer of the YoloV3-slim model) are extracted. These features can capture the high-level semantic information of the target ("head-torso-limbs" structure of a person) and improve the accuracy of category differentiation.

[0312] Feature extraction process logic:

[0313] Feature region localization: Based on the target bounding box, the target region in the image is cropped (excluding background interference);

[0314] Low-level feature extraction: Edge detection and gray-level co-occurrence matrix calculation are performed on the target region to obtain parameters of contour features and texture features;

[0315] High-level feature extraction: Input the target region into the intermediate layer of the AI ​​model to obtain semantic feature vectors;

[0316] Feature fusion: Contour features, texture features, and semantic features are fused according to preset weights to form a "target feature vector", which is stored in the "target feature list". Each feature vector is bound to the corresponding target (associated through the target ID).

[0317] Feature validity verification logic: Verify the extraction effect through "feature discrimination test" - calculate the similarity of feature vectors of different categories of targets (people and vehicles). If the similarity is ≤ preset threshold (indicating that the feature can distinguish the category), the feature is deemed valid; if the similarity is too high, readjust the feature extraction parameters (increase the weight of texture features) to ensure the feature's discriminative ability.

[0318] The principle and process logic of target priority determination

[0319] The core of target priority determination is "quantifying the importance of targets based on target characteristics and task requirements," providing a priority basis for subsequent allocation of sub-machine resources. Its process logic adopts a dual mode of "AI automatic determination + manual review" to balance efficiency and accuracy.

[0320] The principles and logic of AI automatic judgment

[0321] AI automatically determines the priority of a target by comparing its features with a pre-defined feature library. Its core principle is similarity measurement in pattern recognition.

[0322] Feature library construction logic: The pre-defined feature library serves as the benchmark for priority determination and contains "standard features" of different target categories.

[0323] "Vector" - by collecting a large number of labeled samples (1000 personnel samples and 1000 vehicle samples), the feature vector of each sample is extracted, and the average feature vector of samples of the same type is calculated as the standard feature vector of the category; at the same time, a priority is set for each category (personnel and key equipment are high priority, ordinary vehicles are medium priority, and non-key targets are low priority).

[0324] Similarity calculation logic: For each target's feature vector, calculate the "cosine similarity" (which measures the directional consistency between two vectors; the closer the value is to 1, the higher the similarity) with the standard feature vectors of all categories in the feature library.

[0325] If the similarity is greater than or equal to the high priority threshold: it is determined to be a high priority target (personnel, key equipment). This threshold is set based on "false positive rate control of high priority targets" - through sample testing, it is ensured that the false positive rate of high priority targets under this threshold is ≤5%;

[0326] If the similarity is greater than or equal to the medium priority threshold and less than the high priority threshold: it is determined to be a medium priority target (ordinary vehicle). This threshold is set based on "false negative rate control of medium priority targets" - ensuring that the false negative rate of medium priority targets is ≤ 10% under this threshold;

[0327] If the similarity is less than the medium priority threshold, it is determined to be a low priority target (non-critical target, such as trees or rocks).

[0328] Priority adjustment logic: The motion state of a target affects its priority (medium-priority targets moving at high speeds may need to be prioritized). Therefore, a "motion state weight factor" is introduced. For medium-priority targets whose speed exceeds a preset threshold, their similarity is appropriately increased (by adding 0.03 similarity compensation). If the similarity after compensation reaches the high-priority threshold, it is adjusted to high priority. For stationary low-priority targets, their similarity is appropriately reduced to avoid misjudging them as medium priority.

[0329] The principles and logic of manual review

[0330] When the similarity automatically determined by AI is lower than the medium priority threshold (or falls within the ambiguous range), manual review is required. The core purpose is to "correct the AI's uncertain judgment" and avoid misjudgments caused by low feature similarity (target occlusion, abnormal pose).

[0331] Review trigger logic: Set a "review threshold" (below the medium priority threshold). When the similarity of the target is less than the review threshold, the uncertainty of the AI ​​judgment result is high (it is difficult to distinguish whether the suspected target is a person or a piece of debris). The system will automatically trigger manual review - send a "manual review prompt" to the ground terminal APP, and push a magnified image of the target (using electronic zoom technology to magnify the target area to facilitate the operator to observe details).

[0332] Review process logic:

[0333] Image magnification: "Electronic zoom" technology is used to magnify the target area by 3 times. The principle is to expand the pixels of the target area through interpolation algorithm, preserve detailed features (the outline of a person's head, the texture of clothing), and avoid blurring of the image after magnification.

[0334] Detailed annotation: The app automatically annotates key areas of the target (boundary boxes, suspected feature points), prompting operators to pay attention to core details;

[0335] Manual judgment: Based on the magnified image details and the requirements of the mission scenario (in the defense interception scenario, personnel need to be identified first), the operator manually determines the target category and priority, and feeds the results back to the system;

[0336] Result Update: Based on the human judgment results, the system updates the priority information of the target and adds the feature vector of the target to the "human-annotated sample library" for subsequent AI model iteration and optimization (to improve the accuracy of automatic judgment of similar targets).

[0337] The logic for ensuring review efficiency is as follows: To avoid the impact of manual review on real-time performance, a "review time limit" is set - operators must complete the review within a preset time (3 seconds). If there is no response after the time limit, the system will retain the initial AI judgment result by default (and mark it as "pending subsequent confirmation") to ensure that the process is not interrupted. At the same time, for multiple targets that need to be reviewed, the APP sorts them by "similarity from high to low" (the higher the similarity, the lower the uncertainty, and targets with low similarity are reviewed first) to improve review efficiency.

[0338] Priority determination result output logic

[0339] After priority determination, the system generates a "target priority list", which includes the ID, category, priority, motion status, and coordinate information of each target. This list is synchronized to the ground terminal APP and the slave resource scheduling module through the MESH link, providing basic data for slave resource calculation in step 102. At the same time, the APP displays the target distribution in a visual way (marking the target location on the map and using different colors to distinguish the priority: red = high priority, yellow = medium priority, blue = low priority), so that the operator can intuitively grasp the target situation.

[0340] 2. Step 102: Principles and Process Logic of Sub-machine Resource Scheduling Calculation

[0341] The core of slave machine resource scheduling calculation is "determining the number and type of slave machines required based on target needs," ensuring that slave machine resources can cover all targets while avoiding resource waste. Its process logic follows a progressive design of "demand quantification → redundancy coefficient determination → resource adaptation," with the core being the balance between "coverage" and "resource utilization."

[0342] The principle and logic for determining the preset multiple K (redundancy factor)

[0343] The preset multiplier K is a mapping coefficient between the number of targets and the number of sub-machines. Its core function is to introduce resource redundancy to cope with unexpected situations such as target misses or sub-machine failures. The value of K is determined based on the characteristics of the task scenario and experimental data support to ensure that "maximum coverage and minimum resource waste" are achieved in different scenarios.

[0344] The influence of scene characteristics on the K value

[0345] The degree of target threat and distribution characteristics vary significantly across different task scenarios, directly determining the design of the K value:

[0346] Defense interception scenario: The target is "high threat, high mobility" (enemy personnel, vehicles), and the consequences of failing to intercept it are severe (potentially leading to mission failure). Therefore, a high K value is required—resource redundancy is used to improve the interception success rate; simultaneously,

[0347] The target moves quickly, requiring more submachines to achieve multi-directional containment and prevent the target from escaping.

[0348] Security and prevention scenarios: The targets pose a low threat (civilian vehicles, pedestrians), and the consequences of missing them are relatively minor, so the K value can be appropriately reduced to minimize resource waste; at the same time, the targets are relatively concentrated, and coverage can be achieved without excessive redundancy.

[0349] Emergency detection scenario: Targets are scattered (trapped people at disaster sites), requiring a high K value to ensure coverage of all scattered targets; however, the targets have low mobility, so the K value does not need to reach the level of the defense interception scenario, balancing coverage and resources.

[0350] K-value optimization logic supported by experimental data

[0351] The specific value of K is determined through a "controlled variable method experiment," the core of which is to test the "key performance indicators" (target escape rate, resource waste rate) under different K values ​​and select the optimal value.

[0352] Experimental design logic:

[0353] Control variables: fixed task scenario (defense interception), number of targets (10), submachine performance, only change the K value (from 1.0 to 2.5);

[0354] Metrics measurement: For each K value, repeat the experiment 100 times and measure the target escape rate (the proportion of targets that are not intercepted) and the resource waste rate (the proportion of submachines that do not participate in the interception).

[0355] Optimal K value selection logic:

[0356] Defense interception scenario: When K=2.0, the target escape rate decreases from 15% when K=1.0 to below 5% (meets mission requirements), and the resource waste rate is 20% (within an acceptable range); if K>2.0, the escape rate decreases insignificantly (<1%), but the resource waste rate increases significantly (>30%), therefore K=2.0 is determined;

[0357] Security and control scenario: When K = 1.5, the target escape rate is 8% (meets the requirements), and the resource waste rate is 10% (lower than 30% when K = 1.0), therefore K = 1.5 is determined;

[0358] Emergency detection scenario: When K=1.8, the target coverage (the proportion of detected targets) increases from 80% when K=1.5 to over 95%, and the resource waste rate is 15%. Therefore, K=1.8 is determined.

[0359] K-value dynamic adjustment logic: The system supports adjusting the K-value according to the real-time target situation. If the number of targets suddenly increases (5 new high-priority targets are added), the K-value will be automatically increased by 0.2 (from 2.0 to 2.2) to ensure resource coverage. If the number of targets decreases or most targets are intercepted, the K-value will be automatically reduced to reduce resource waste.

[0360] The principle and process logic for calculating the total number of submachines and the proportion of different types

[0361] The core of calculating the total number of sub-machines is "determining the minimum number of sub-machines based on the target number and the K value"; the core of calculating the type ratio is "determining the ratio of high-speed racing drones to AI racing drones based on the target priority distribution" to ensure that the sub-machine types match the target requirements.

[0362] The principle and logic of calculating the total number of slave machines

[0363] The total number of slave machines is calculated using the formula "M=ceil(N×K)", where the "ceil function" (rounding up) is crucial.

[0364] Its principles and logic are as follows:

[0365] Formula design logic:

[0366] N is the target quantity (from the target priority list in step 101), K is the preset multiple, and N×K is the "theoretical number of submachines required" (including redundancy).

[0367] The purpose of the ceil function is to "avoid decimal submachines"—the number of submachines must be an integer. Rounding up ensures that the number of submachines is sufficient to cover the theoretical requirement (N=5, K=2.0, N×K=10.0, after ceil M=10; if N=6, K=2.0, N×K=12.0, M=12); if rounding down is used, the number of submachines may be insufficient (N=5, K=2.0, rounded down to 9, which may not cover the theoretical requirement of 10 submachines).

[0368] Total number verification logic: After calculating M, it is necessary to verify whether the "minimum coverage requirement" is met - through the "target-submachine ratio test", ensure that each target corresponds to at least 1.2 submachines (based on redundancy requirements). If M / N < 1.2, the K value needs to be readjusted (appropriately increased) to ensure coverage. At the same time, verify whether M exceeds the system's maximum submachine capacity (master unit load + transmitter capacity). If it exceeds, take the system's maximum capacity as the actual total number of submachines (and mark it as "insufficient resources" to prompt the operator to prioritize the allocation of high-priority targets).

[0369] The principle and logic of calculating the proportion of sub-machine types

[0370] Sub-machine types are divided into "high-speed racing drones" (advantages: fast response, suitable for high-priority targets) and "AI racing drones" (advantages: high accuracy, suitable for medium- and low-priority targets). The type ratio is designed based on "target priority distribution + sub-machines".

[0371] "Machine performance matching" is based on the following principle:

[0372] Design logic for percentage constraints:

[0373] High-speed racing drones account for ≥40% of the total: High-priority targets (personnel) require rapid response. The flight speed of high-speed racing drones is more than 1.5 times that of AI racing drones, enabling them to quickly reach the target area. By setting a ≥40% proportion, it is ensured that there are enough high-speed drones to deal with high-priority targets (among 10 drones, at least 4 are high-speed drones, which can cover 3-4 high-priority targets).

[0374] AI-powered racing drones should account for ≤60% of the total: For low-to-medium priority targets (ordinary vehicles), precise operations (precise detection and marking) are required. The AI ​​recognition accuracy of AI-powered racing drones is higher than that of high-speed racing drones. Setting the percentage to ≤60% avoids resource waste caused by too many AI sub-drones (high-priority targets do not require high precision, and high-speed sub-drones are more efficient).

[0375] Percentage calculation process logic:

[0376] Priority distribution statistics: Count the number of high-priority and medium-low-priority targets in the target priority list (3 high-priority targets and 7 medium-low-priority targets).

[0377] Minimum number of high-speed submachines calculated: Number of high-speed submachines ≥ max(40% × M, number of high-priority targets) – This satisfies both the percentage constraint and ensures that each high-priority target corresponds to at least one high-speed submachine (to avoid high-priority targets being without a corresponding high-speed submachine).

[0378] (High-speed slave unit matching)

[0379] AI Sub-machine Count Calculation: AI Sub-machine Count = M - High-Speed ​​Sub-machine Count, and the AI ​​Sub-machine Count must be ≤ 60% × M (if it exceeds this, appropriately reduce the AI ​​Sub-machine Count and increase the High-Speed ​​Sub-machine Count to ensure compliance with the percentage constraint).

[0380] Hardware capacity adaptation logic: The calculated number of high-speed slave units and AI slave units must match the system hardware capacity (maximum number of high-speed slave units that the launcher can accommodate, maximum number of AI slave units that the main unit can mount).

[0381] If the number of high-speed sub-units is greater than the launcher capacity: the launcher capacity is taken as the actual number of high-speed sub-units, and the remaining high-speed sub-unit demand is supplemented through "standby sub-unit scheduling" (calling the standby high-speed sub-units reserved by the system).

[0382] If the number of AI slave machines is greater than the capacity of the host machine: take the capacity of the host machine as the actual number of AI slave machines, and solve the remaining AI slave machine requirements through "task splitting" (prioritize the allocation of AI slave machines to medium-priority targets, and replace low-priority targets with backup AI slave machines or high-speed slave machines).

[0383] Backup slave machine setting: If the actual total number of slave machines is less than the maximum system capacity, the remaining slave machines are set as "backup slave machines" - high-speed slave machines are given priority as backups (because high-speed slave machines respond quickly and are suitable for filling in missed targets). The number of backup slave machines = maximum system capacity - total number of actual slave machines, to ensure that resources are available in case of emergencies.

[0384] Resource scheduling result output logic

[0385] After the sub-machine resource scheduling calculation is completed, the system generates a "sub-machine resource scheduling scheme", which includes:

[0386] Actual total number of slave units: M_actual (matching hardware capacity);

[0387] Number of high-speed racing drones: M_high-speed (meeting ≥40% percentage);

[0388] Number of AI racing drones: M_AI (meeting ≤60% percentage);

[0389] Number and type of backup sub-units: M_Backup (priority high-speed sub-units);

[0390] Resource adaptation instructions: If there is insufficient resources (the number of high-speed sub-machines does not meet the requirements), please explain the alternative solution (use AI sub-machines as a substitute, and the matching strategy needs to be adjusted).

[0391] The solution is synchronized to the ground terminal APP and the target-sub-machine matching module, providing a resource basis for the matching and allocation in step 103; the APP displays the resource distribution in the form of charts (a pie chart showing the proportion of high-speed sub-machines, AI sub-machines, and backup sub-machines), making it easy for operators to confirm whether the resources meet the task requirements.

[0392] 3. Step 103: Principles and process logic of target-sub-machine matching and allocation

[0393] The core of target-sub-machine matching and allocation is "to achieve the optimal match between target requirements and sub-machine capabilities through algorithm optimization," ensuring that high-priority targets are assigned to the most suitable sub-machines while minimizing the sub-machine's flight costs (distance, energy consumption). Key distance constraints: The AI ​​racing drone and the mother drone are deployed 1 km above the target (the straight-line distance D from the AI ​​sub-machine to the target is 1 km); the straight-line distance between the launch pad of the high-speed assault racing drone and the mother drone / AI racing drone is 15 km, meaning the straight-line distance D from the high-speed sub-machine to the target is 15 km. Its process logic is based on "matching matrix construction → improved Hungarian algorithm optimization..."

[0394] "Quantification → Result Verification and Confirmation" is essentially about balancing "cost minimization" and "demand maximization".

[0395] The principles and logic of matching matrix construction

[0396] The matching matrix is ​​a quantitative representation of the "adaptation relationship" between the target and the sub-machine. Its core is to define the "adaptation cost" (the lower the cost, the stronger the adaptability). The design of the matrix elements is based on "performance matching degree + distance matching degree". Combined with the distance characteristics of 15KM for high-speed sub-machine and 1KM for AI sub-machine, it ensures that the matching result meets the target requirements while taking into account the efficiency of the sub-machine.

[0397] (I) Quantitative Principles of Performance Matching

[0398] Performance matching degree is "the degree to which the submachine's capabilities match the target requirements". The quantification logic is based on "the correspondence between submachine type and target priority", as follows:

[0399] Correspondence design logic:

[0400] The core requirement for high-priority targets (personnel) is "rapid response." High-speed racing drones fly at speeds more than 1.5 times faster than AI racing drones, meeting the need for rapid interception. Therefore, the performance match between high-priority targets and high-speed racing drones is 1.0 (highest, strongest compatibility); the performance match between high-priority targets and AI racing drones is 0.6 (lower, weaker compatibility) – the slow response of AI racing drones may lead to target escape. The core requirement for medium-priority targets (ordinary vehicles) is "precise operation" (precise detection and marking). The AI ​​recognition accuracy of AI racing drones (≥95%) is higher than that of high-speed racing drones (≥85%). Therefore, the performance match between medium-priority targets and AI racing drones is 1.0; the performance match between medium-priority targets and high-speed racing drones is 0.7 (lower) – the insufficient accuracy of high-speed racing drones may lead to operational errors. Low-priority targets (non-critical targets) have lower requirements, and their performance match with both types of sub-machines is set to 0.8 (moderate compatibility). There is no need to strictly distinguish between sub-machine types; idle sub-machine resources should be utilized first.

[0401] The conversion logic between matching degree and cost

[0402] Adaptation cost is negatively correlated with performance matching degree; that is, the higher the performance matching degree, the lower the adaptation cost (cost = 1 - performance matching degree). For example:

[0403] High-priority target + high-speed racing drone: performance matching degree = 1.0 → adaptation cost = 0;

[0404] High-priority target + AI racing drone: Performance matching degree = 0.6 → Adaptation cost = 0.4;

[0405] Medium-priority target + AI racing drone: Performance matching degree = 1.0 → Adaptation cost = 0;

[0406] Medium-priority target + high-speed racing drone: performance matching degree = 0.7 → adaptation cost = 0.3;

[0407] Low-priority target + arbitrary sub-machine: performance matching degree = 0.8 → adaptation cost = 0.2.

[0408] This conversion logic ensures that during the optimization process, the algorithm prioritizes the target-submachine combination with high performance matching, which meets the task requirement of "using the optimal submachine for high-priority targets".

[0409] (II) Calculation principle of distance matching degree

[0410] Distance matching degree is a quantitative indicator of "flight efficiency from deployment of the sub-unit to the target." The shorter the flight distance, the shorter the time it takes for the sub-unit to reach the target (faster response) and the lower the energy consumption (longer endurance). Therefore, the higher the distance matching degree, the lower the adaptation cost. The calculation logic based on key distance constraints is as follows:

[0411] Flight distance prediction logic

[0412] Submachine deployment location determined:

[0413] High-speed racing drone: The mother drone / AI racing drone is 15km above the target from the launch pad → the straight-line distance D from the high-speed drone to the target is 15km;

[0414] AI racing drone: Dropped from the mother drone (mother drone is 1KM above the target) → AI drone to the target in a straight line, distance D = 1KM;

[0415] Target coordinate acquisition: Extract the real-time coordinates of the target from the target priority list in step 101;

[0416] Straight-line distance calculation: Ignoring terrain obstruction (open outdoor scene), the distance from the high-speed sub-machine to the target is fixed at 15KM, and the distance from the AI ​​sub-machine to the target is fixed at 1KM, as an approximation of the flight distance (the actual flight distance increases slightly due to obstacle avoidance, but the straight-line distance can meet the matching accuracy requirements).

[0417] 2. Quantification logic of distance matching degree (based on D0 = 15 km threshold)

[0418] Distance threshold setting: Based on the maximum effective operating radius of the sub-machine, the distance threshold D0 is set to 15 kilometers. When D≤D0, the sub-machine can quickly reach the target; when D>D0, the sub-machine response time is too long and the adaptability is reduced.

[0419] Matching degree calculation: Distance matching degree is quantified using a "linearly decreasing function".

[0420] If D≤D0: Distance matching degree = 1-(D / D0)×0.1;

[0421] High-speed slave unit (D=15KM): Distance matching degree = 1-(15 / 15)×0.1=0.9;

[0422] AI sub-machine (D= 1KM): Distance matching degree = 1 - (1 / 15) × 0.1 ≈ 0.9933;

[0423] If D>D0: Distance matching degree = 0.9-(D-D0) / D0×0.4.

[0424] 3. The conversion logic of distance cost

[0425] Distance cost = 1 - distance matching degree, ensuring that the closer the target and sub-machine combination, the lower the cost:

[0426] High-speed slave unit (D=15KM): Distance cost = 1-0.9=0.1;

[0427] AI submachine (D= 1KM): Distance cost = 1 - 0.9933 ≈ 0.0067;

[0428] (III) Calculation of total adaptation cost and matrix construction

[0429] The total adaptation cost is a weighted sum of performance cost and distance cost. The weights are set based on the principle of "task priority over efficiency"—performance cost weight W1 = 0.6, distance cost weight W2 = 0.4; total cost calculation: total adaptation cost C = W1 × performance cost + W2 × distance cost; matrix construction: let the number of targets be N and the number of sub-machines be M, construct an N-row M-column matching matrix M [N×M], where matrix element M[i][j] = the total adaptation cost between target i and sub-machine j.

[0430] Target 1 (high priority) and Sub-machine 1 (high speed, D= 15KM): Performance cost = 0, distance cost = 0.1 → C=0×0.6+0.1×0.4=0.04;

[0431] Target 1 (high priority) and Sub-machine 2 (AI, D= 1KM): Performance cost = 0.4, distance cost ≈ 0.0067 → C = 0.4×0.6 + 0.0067×0.4 ≈ 0.2427;

[0432] Target 2 (medium priority) and Sub-machine 2 (AI, D= 1KM): Performance cost = 0, distance cost ≈ 0.0067 → C=0×0.6+0.0067×0.4≈0.0027;

[0433] Target 2 (medium priority) and Sub-machine 1 (high speed, D= 15KM): Performance cost = 0.3, distance cost = 0.1 → C=0.3×0.6+0.1×0.4=0.22;

[0434] Target 3 (low priority) and Sub-machine 1 (high speed, D= 15KM): Performance cost = 0.2, distance cost = 0.1 → C=0.2×0.6+0.1×0.4=0.16;

[0435] Target 3 (low priority) and Sub-machine 2 (AI, D= 1KM): Performance cost = 0.2, distance cost ≈ 0.0067 → C = 0.2×0.6 + 0.0067×0.4 ≈ 0.1227.

[0436] The matching matrix fully presents the adaptation cost of all target-submachine combinations, providing basic data for subsequent optimization calculations of the Hungarian algorithm.

[0437] The principle and logic of the improved Hungarian algorithm for optimal matching

[0438] The traditional Hungarian algorithm is suitable for one-to-one matching scenarios where "number of targets = number of sub-machines". However, in this system, the number of targets and the number of sub-machines may not be equal (sub-machines include redundancy), and the matching needs of high-priority targets must be prioritized. Combining the fixed distance characteristics of high-speed sub-machines (15KM) and AI sub-machines (1KM), the core logic of the improved Hungarian algorithm is "staged matching + priority constraint + distance cost weight" to ensure that the matching result satisfies both the minimization of total cost and the requirements of task priority.

[0439] (I) Core Ideas for Algorithm Improvement

[0440] Phased matching: The matching process is divided into three phases: "high-priority target matching → medium-priority target matching → low-priority target matching". Each phase only uses the corresponding adapted sub-machine resources to avoid sub-machine resources being occupied by low-priority targets.

[0441] Priority constraints: The matching cost threshold for high-priority targets is lower than that for medium- and low-priority targets (maximum acceptable cost for high-priority targets = 0.3, medium priority = 0.5, low priority = 0.8). The total cost of a high-speed submachine for a high-priority target (0.04) is much lower than the threshold, so it is matched first. If the cost of a high-priority target with all available submachines exceeds the threshold, "emergency resource scheduling" (calling a backup high-speed submachine) is triggered.

[0442] Distance cost weighting: High-speed sub-machine distance cost 0.1, AI sub-machine distance cost ≈ 0.0067, in total cost calculation.

[0443] This does not affect the core logic of "high-priority targets are matched with high-speed sub-machines first, and medium-priority targets are matched with AI sub-machines first";

[0444] Redundant Sub-machine Allocation: After all target matching is completed, the remaining sub-machines are set as "backup sub-machines for supplementary interception" and are allocated to the matched targets according to the "nearest distance principle" (for AI sub-machines, due to the short distance of 1KM, the backup priority is higher than that of high-speed sub-machines; AI sub-machines are preferentially allocated to medium-priority targets at a distance of 1KM, and high-speed sub-machines are preferentially allocated to high-priority targets at a distance of 15KM).

[0445] (2) Specific Process Logic for Phased Matching

[0446] Phase 1: Matching of High-Priority Targets and High-Speed Sub-machines (D = 15KM)

[0447] Resource Screening: Screen out all high-speed sub-machines (quantity M1) from the sub-machine resources, and screen out high-priority targets (quantity N1) from the target list;

[0448] Judgment of the Quantity of Sub-machines

[0449] If M1 ≥ N1: Construct a sub-matrix with N1 rows and M1 columns (only including the costs of high-priority targets and high-speed sub-machines, fixed distance cost 0.1), execute the improved Hungarian algorithm to achieve one-to-one matching of N1 targets and N1 high-speed sub-machines (the total cost is at least 0.04), and the remaining M1 - N1 high-speed sub-machines enter the "backup sub-machine pool";

[0450] If M1 < N1: First, match M1 high-speed sub-machines with M1 high-priority targets (with the optimal total cost based on the 15KM distance cost), and the remaining N1 - M1 high-priority targets enter the "pending matching pool". Subsequently, call AI sub-machines (the total cost ≈ 0.2427, still lower than the high-priority threshold of 0.3), and it is necessary to manually confirm whether to accept the response delay risk of AI sub-machines.

[0451] Phase 2: Matching of Medium-Priority Targets and AI Sub-machines (D = 1KM)

[0452] Resource Screening: Screen out all AI sub-machines (quantity M2) from the sub-machine resources, and screen out medium-priority targets (quantity N2) from the target list;

[0453] Judgment of the Quantity of Sub-machines

[0454] If M2 ≥ N2: Construct a sub-matrix with N2 rows and M2 columns (only including the costs of medium-priority targets and AI sub-machines, D = 1KM → total cost ≈ 0.0027), execute the improved Hungarian algorithm to achieve one-to-one matching of N2 targets and N2 AI sub-machines, and the remaining M2 - N2 AI sub-machines enter the "backup sub-machine pool";

[0455] If M2 < N2: First, match M2 AI sub-machines with M2 medium-priority targets. For the remaining N2 - M2 medium-priority targets, call the high-speed sub-machines (total cost 0.22, lower than the medium-priority threshold of 0.5). It is necessary to manually confirm whether to accept the risk of insufficient accuracy of the high-speed sub-machines.

[0456] Phase 3: Matching of low-priority targets with the remaining sub-machines

[0457] Resource screening: Screen out the remaining sub-machines (high-speed + AI, with a quantity of M3) from the "backup sub-machine pool", and screen out low-priority targets (with a quantity of N3) from the target

[0458] list;

[0459] Matching logic: Construct a sub-matrix with N3 rows and M3 columns (including the 15KM distance cost of high-speed sub-machines and the 1KM distance cost of AI sub-machines), execute the improved Hungarian algorithm, and preferentially match AI sub-machines (total cost ≈ 0.1227 < 0.16 for high-speed sub-machines). The remaining sub-machines remain in the "backup sub-machine pool" as global backup for blocking.

[0460] III. Instruction Issuance and Sub-machine Execution Phase

[0461] In the instruction issuance and sub-machine execution phase, it is necessary to complete the reliable transmission of instructions and the precise deployment of sub-machines, ensure the stability of data interaction through dual-link coordination, and ensure that sub-machines execute according to task requirements.

[0462] 1. Generation and Transmission of Instructions <F

[0463] The ground terminal generates a scheduling instruction based on the matching result and transmits it to the mother machine / launch rack through a stable link:

[0464] Structured Design of Instructions

[0465] The scheduling instruction is encapsulated using the MAVLinkv2 protocol and contains the following fields:

[0466] Instruction header: Includes instruction type (launch / toss), target ID;

[0467] Task parameters: Include target coordinates (in WGS84 format), end-guidance start distance (500 meters for high-speed drones, 400 meters for AI drones);

[0468] Mode parameters: Include launch / toss mode (batch / precise), interval time (1 second for precise toss);

[0469] Checksum: Adopt CRC32 checksum to ensure the integrity of instruction transmission.

[0470] Command link transmission

[0471] Commands are transmitted via a dual-link system (wired + wireless) to ensure reliability.

[0472] Transmitter commands: transmitted to the transmitter via gigabit network cable (wired link), with a transmission rate ≥10Mbps and a command transmission delay ≤50ms;

[0473] Master machine command: Transmitted to the master machine via MESH wireless link. The command is marked as "high priority" and occupies 30% of the link bandwidth to avoid being occupied by non-critical data.

[0474] 2. Step 104: Batch Launch

[0475] After the ground terminal sends a "batch launch command" to the launcher, the launcher executes the following procedure:

[0476] Pre-launch preparations

[0477] Magnetic interface data pre-transmission: The launcher pre-transmits mission parameters to three high-speed racing drones via a magnetic interface, with a data transmission rate ≥10Mbps and a pre-transmission duration ≤0.3 seconds;

[0478] Start-up signal trigger: After the pre-transmission is completed, the magnetic interface sends a high-level start signal (3.3V), and the flight control system of the high-speed racing drone starts the motor self-test.

[0479] Launch execution

[0480] Automatic lid opening: The DC motor of the launcher drives the lid to open. The motor uses PWM speed regulation (duty cycle 80%), and the opening time is ≤3 seconds.

[0481] Rail-guided launch: The electromagnetic latches of the three high-speed racing drones are unlocked simultaneously, and they are launched vertically through the rail mechanism. The guiding accuracy of the rail is ≤1°, ensuring the accuracy of the initial flight direction of the drones.

[0482] Status feedback: After launch, the launcher sends a "launch successful" signal back to the ground terminal, including the slave unit ID and launch time.

[0483] 3. Step 104: Precise Throw

[0484] After the ground terminal issues a "precision delivery command" to the mothership, the mothership executes the following procedures:

[0485] Mother machine hovering attitude adjustment

[0486] The machine machine adjusts to a stable hovering state through a PID attitude control algorithm:

[0487] Roll / Pitch Angle Adjustment: The flight control system collects IMU data in real time and adjusts the motor speed to ensure that the roll / pitch angle deviation is ≤1°;

[0488] Height Maintenance: The motor speed is adjusted based on barometer data to ensure a height deviation of ≤0.5 meters, guaranteeing throwing stability. Precise Throwing Execution

[0489] Magnetic interface data pre-transmission: The host machine pre-transmits task parameters to the AI ​​racing drone via the magnetic interface, with a pre-transmission time of ≤0.2 seconds;

[0490] Dropping drones one by one: The electromagnetic locks unlock sequentially at 1-second intervals, allowing the AI ​​racing drone to detach from the mother machine. The motors start 0.1 seconds after detachment to prevent mid-air collisions.

[0491] Status feedback: After each AI racing drone completes its drop, it sends a "drop successful" signal back to the mothership, which then summarizes and forwards it to the ground terminal.

[0492] IV. Principles and Process Logic of Status Feedback and Interception Phase

[0493] The status feedback and interception phase is a crucial link in the completion of the mission loop of the mother-daughter UAV swarm system. It follows the scheduling scheme of the sub-drone in the "target detection and allocation phase" and ensures the integrity of target interception and the traceability of mission results through the whole process logic of "precise interception execution → real-time missed detection → dynamic interception response → mission completion and archiving".

[0494] This phase follows three principles: "real-time priority, closed-loop control, and resource optimization." "Real-time priority" means ensuring timely detection and handling of missed targets through high-frequency data interaction and rapid command response. "Closed-loop control" means that each interception action includes a closed loop of "execution → feedback → judgment → adjustment" to avoid interception failures without follow-up measures. "Resource optimization" means prioritizing the use of idle and efficient resources during supplementary interception scheduling to reduce slave unit losses and energy waste. The principles and process logic of each stage are explained in detail below with reference to steps 105-106 in the attached diagram.

[0495] 1. Step 105: The principle and process logic of target interception execution

[0496] Target interception and execution is the core step in enabling sub-machines to implement precise control over targets according to the matching scheme. This requires consideration of the sub-machine type.

[0497] The performance differences between high-speed racing drones and AI racing drones necessitate the design of differentiated guidance and tracking strategies to ensure the accuracy and reliability of interception actions. The core logic is "adapting guidance schemes to sub-drones' capabilities → dynamically adjusting trajectories → determining interception effectiveness," with each sub-drone's interception process forming an independent control closed loop.

[0498] The principle and logic of dual-mode guidance for high-speed racing drones

[0499] The core advantages of high-speed racing drones are their high flight speed and rapid response, which meet the needs of rapid interception of high-priority targets. The design principle of its dual-mode guidance is "multi-source data fusion to improve positioning accuracy → dynamic trajectory adjustment based on precise positioning". By complementing the advantages of radar and vision, the accuracy defects of single guidance methods are solved.

[0500] Cooperative logic of dual-mode guidance:

[0501] The core function of radar detection is "long-range coarse positioning": radar has strong anti-jamming capabilities (unaffected by light or weather) and can obtain the relative position and speed of the target at a relatively long distance, providing initial flight direction guidance for the sub-aircraft; its logical defect is that the positioning accuracy at close range is insufficient, which cannot meet the requirements of precise terminal interception.

[0502] The core function of visual recognition is "precise positioning at close range": visual sensors can capture the contour details of the target and output high-precision target bounding box coordinates to make up for the short-range accuracy defects of radar; its logical defects are that it is susceptible to sudden changes in lighting and target occlusion, and has poor stability in long-range recognition.

[0503] Cooperative mechanism: When the high-speed sub-machine flies to the set terminal guidance distance, the radar and vision are activated simultaneously. The radar provides global motion parameters (velocity, long-range position) and the vision provides local detailed parameters (boundary box, close-range position). The two types of data are integrated through data fusion algorithm to form a target positioning result that has both anti-interference ability and high accuracy.

[0504] The closed-loop logic of data fusion:

[0505] The core principle of the fusion algorithm is to use the Kalman filter algorithm, taking radar data and visual data as separate observation inputs to the system, and eliminating noise interference and errors from individual data through an iterative "prediction → update" process. Specifically, the logic is as follows: first, based on the historical motion states of the sub-machine and the target, the target position at the current moment is predicted (prediction step); then, the real-time observation data from radar and vision are compared with the predicted value, the error is calculated, and the prediction result is corrected (update step); finally, the fused, accurate positioning data is output.

[0506] The verification logic for the fusion effect is as follows: During the fusion process, the stability of the positioning data (the fluctuation range of the positioning results of 5 consecutive frames) is calculated in real time. If the fluctuation range is less than or equal to the preset threshold (indicating that the positioning is stable), the fused data is adopted. If the fluctuation range exceeds the threshold, a certain data source is determined to be abnormal (visual obstruction), and the weight of another data source is automatically increased (the weight of radar data is increased) to ensure the reliability of the fusion result.

[0507] Trajectory Adjustment and Interception Judgment Logic:

[0508] The core logic of trajectory adjustment is as follows: the flight control system uses the fused target positioning data as control input, compares it with the current position of the slave aircraft, and calculates the position deviation; based on the deviation value, the flight attitude (roll, pitch, speed) of the slave aircraft is adjusted through a PID control algorithm.

[0509] By adjusting the pitch angle and motor output power, the position deviation is gradually reduced, allowing the submachine to approach the target.

[0510] The logic for interception determination is as follows: when the distance between the sub-machine and the target is reduced to a set threshold (the minimum distance to ensure the interception effect), and the target is within the effective control range of the sub-machine (the visual bounding box completely covers the target), the interception is determined to be successful; if the target's motion state changes suddenly during the approach (sudden acceleration), data fusion and trajectory adjustment are re-executed until the interception determination conditions are met or a missed detection warning is triggered.

[0511] The Principles and Logic of AI-Powered Racing Drones for Precise Tracking and Interception

[0512] The core advantages of AI racing drones are high visual recognition accuracy and strong tracking stability, which are suitable for the precise control needs of low-to-medium priority targets. Its interception logic is based on "SiamRPN tracking algorithm → dynamic template update → IOU judgment interception", which ensures that the target can still be stably tracked and intercepted when it is moving or changing its posture.

[0513] The core principle of tracking algorithms:

[0514] The logical advantages of the SiamRPN algorithm are as follows: This algorithm achieves real-time target tracking through a two-branch structure of a "template branch" and a "search branch." The template branch extracts the initial feature template of the target and stores the target's key visual information (contour, texture); the search branch extracts image features within the sub-machine's field of view in real time, performs similarity matching with the template features, and outputs candidate positions of the target. Compared to traditional tracking algorithms, its advantage lies in its stronger adaptability to changes in target scale and rotation, and its reduced likelihood of tracking loss due to changes in target pose.

[0515] The logic for initializing the tracking box is as follows: the target bounding box obtained in the "target detection and allocation stage" is used as the initial tracking box. The target image within the box is input into the template branch to generate the initial feature template. At the same time, the update threshold of the tracking box is set (the position correction is triggered when the target position offset exceeds 20% of the tracking box area) to ensure that the tracking box always surrounds the target.

[0516] Dynamic logic for template updates:

[0517] Necessity of Template Updates: Targets may change appearance during movement (vehicles turning to reveal their sides, people changing posture). Using the initial template indefinitely will lead to decreased similarity matching accuracy and tracking loss. Therefore, templates need to be updated periodically to ensure their features remain consistent with the target's current appearance.

[0518] Update cycle and strategy: A fixed template update cycle is set (updated every 5 frames). During the update, the target image within the current tracking box is re-input into the template branch to generate a new feature template. At the same time, a "similarity verification" mechanism is introduced. If the similarity between the new template and the old template is greater than or equal to a preset threshold (indicating that the target appearance has changed little), the old template is replaced. If the similarity is too low (indicating a sudden change in the target appearance, which may be due to occlusion or mistracking), the template update is paused, and the target is reconfirmed through visual recognition to avoid tracking deviation caused by incorrect templates.

[0519] The logic for determining whether to intercept execution is as follows:

[0520] The principle of IOU determination: IOU (Intersection over Union) is an indicator that measures the degree of overlap between the tracking box and the actual position of the target. The closer the IOU value is to 1, the more complete the coverage of the target by the tracking box. When IOU ≥ a set threshold, it indicates that the slave device has successfully tracked the target.

[0521] It has been brought under effective control and is capable of being intercepted.

[0522] Trajectory Adjustment and Interception Execution: The flight control system adjusts the flight trajectory of the slave aircraft based on the positional deviation of the tracking frame (the deviation between the center of the tracking frame and the center of the slave aircraft's field of view) to keep the target always in the center of the tracking frame; when the IOU reaches the interception threshold and the distance between the slave aircraft and the target is less than or equal to the set value, the interception action is executed; if the IOU remains below the threshold during the tracking process (the target escapes quickly), a missed target warning is triggered, and the system waits for a supplementary interception response.

[0523] 2. Step 106: The principle and process logic of intercepting missed targets.

[0524] Recapturing missed targets is a key guarantee for ensuring mission integrity. Its core logic is "real-time monitoring of interception status → rapid determination of missed targets → dynamic scheduling of optimal submachines → planning of efficient interception paths". Through a closed-loop monitoring and response mechanism, the target miss rate is minimized.

[0525] The principles and logic for identifying targets that slip through the net

[0526] Detecting missed targets is a prerequisite for re-interception. It requires multi-dimensional status monitoring to identify three types of missed targets: "interception failure, sub-machine failure, and new target addition," to ensure no omissions and no misjudgments.

[0527] The logic of multi-dimensional monitoring:

[0528] Tracking status monitoring: The slave unit transmits tracking data (tracking frame IOU value, target position) back to the ground terminal in real time. The ground terminal analyzes the tracking status at fixed intervals (every 1 second). If the tracking data of a target is interrupted (no tracking information is received) or the IOU value remains below the tracking threshold for more than a set time (3 seconds), it is determined that "the target is not being tracked" and a missed target warning is triggered.

[0529] Sub-unit status feedback: If a fault occurs during the interception process (power system malfunction, vision sensor failure), the sub-unit will automatically send back a "mission failed" signal; after receiving the signal, the ground terminal will directly determine that the corresponding target is a target that has slipped through the net without waiting for tracking status monitoring, and trigger a re-interception.

[0530] New target detection: The AI ​​recognition module of the mothership continuously scans the target area during the interception phase of the slave aircraft. If a new target that has not been assigned to a slave aircraft is found (confirmed by comparing it with the list of assigned targets), the new target information is immediately synchronized to the ground terminal, and it is determined to be a target that has slipped through the net, triggering a supplementary interception.

[0531] False positive logic for judgment:

[0532] Multi-source cross-validation: In the case of "target not being tracked", cross-validation is required using the visual monitoring data of the master unit. If the master unit can still identify the target, it is confirmed that the target has slipped through the net; if the master unit cannot identify it (the target has left the monitoring area), then a re-interception will not be triggered to avoid misjudgment caused by the master unit losing tracking but the target having left the field.

[0533] Fault confirmation mechanism: For the "mission failed" signal returned by the slave unit, the ground terminal needs to send a confirmation command. Only after the slave unit reports the fault status again will the target be determined to have escaped the network. If the slave unit recovers and returns a "mission normal" signal, the determination of escaped the network will be cancelled to avoid misjudgment caused by temporary communication interruption.

[0534] Scheduling and path planning logic of the roadblock repair machine

[0535] The core of the interception and blocking scheduling is "quickly calling the optimal slave machine resources", and the core of path planning is "ensuring that the slave machine arrives at the target location that was missed in the shortest time". The combination of the two achieves efficient interception and blocking.

[0536] Scheduling priority logic for the rebar repair machine:

[0537] The principle of priority setting: Scheduling priorities are set based on the "response speed" and "availability" of the sub-machines, ensuring that the fastest available sub-machine is called first. The specific priority sorting logic is as follows:

[0538] First priority: Idle AI drones (no task assigned, in standby mode) - AI sub-machines have fast response speed and are closest to the target, making them suitable for emergency interception;

[0539] Second priority: High-speed racing machine that has just finished charging (has just finished charging and has not yet been assigned a task) - its response speed is slower than that of the AI ​​sub-machine, and it is farther away from the target. It is only used when no AI sub-machine is available.

[0540] Third priority: Idle high-speed racing drone (completed charging, no task assigned, in standby mode) - compared to idle high-speed sub-drones, only a short startup preparation is required, and the response speed is similar;

[0541] The execution logic of the scheduling is as follows: the ground terminal searches for sub-machine resources in order of priority. After finding the first available sub-machine, it immediately generates a replacement interception command. If there is no available sub-machine for a certain priority, it automatically searches for the next priority to ensure the timeliness of the replacement interception command generation (command generation is completed within 100ms).

[0542] The optimization logic of the obstacle replacement path planning:

[0543] The adaptation principle of Dijkstra's algorithm: This algorithm is a typical shortest path algorithm. It calculates the weights of all possible paths from the current position of the slave device to the position of the target that slipped through the net, and selects the path with the smallest weight. In the interception scenario, the path weight needs to take into account both "flight distance" and "target priority" to ensure that high-priority targets that slipped through the net are dealt with first.

[0544] Weight setting and path selection:

[0545] Flight distance weight: Based on the straight-line distance from the current position of the aircraft to the target position, multiplied by a distance coefficient (1.0). The greater the distance, the greater the weight of this part.

[0546] Target priority weight: The priority coefficient for high-priority targets is set to 2.0, and for medium- and low-priority targets it is set to 1.0. The higher the target priority, the greater the weight of this part.

[0547] Total weight calculation: Total path weight = Flight distance × Distance coefficient + Target priority × Priority coefficient. The algorithm compares the total weight of all paths and selects the path with the smallest total weight as the interception path—ensuring both the shortest flight distance for the sub-machine and prioritizing the interception efficiency of high-priority targets.

[0548] Dynamic path adjustment: During the interception process, if the sub-aircraft detects an obstacle (other sub-aircraft, building) in the path, the path weight is recalculated in real time and the flight path is adjusted to avoid collisions and path congestion.

[0549] Closed-loop control logic for supplementary interception execution

[0550] The interception and backup execution is the implementation phase of scheduling and path planning. It requires a closed loop of "instruction issuance → slave response → interception monitoring → result feedback" to ensure successful interception and backup.

[0551] Command issuance and slave response logic:

[0552] Command content: The interception command includes the coordinates of the target that slipped through the net, its priority, the planned path, and the interception method (the high-speed slave unit uses dual-mode guidance) to ensure that the slave unit clearly understands the mission requirements;

[0553] Response mechanism: After receiving the interception command, the slave unit immediately enters the start-up state (the launcher starts the high-speed slave unit's launch program, and the idle slave unit starts the power system) and sends back the "command reception" signal to the ground terminal; if the slave unit fails to send back the signal within the set time (100ms), the ground terminal reissues the command to avoid command loss.

[0554] Interception and result feedback logic:

[0555] Interception process monitoring: The sub-unit flies along the interception path and, upon reaching the terminal guidance distance, executes the same guidance / tracking logic as the initial interception; the ground terminal receives the interception status data of the sub-unit in real time and monitors the interception progress.

[0556] Result judgment and feedback: If the slave unit sends back a "successful interception" signal, or the master unit visually confirms that the target is under control, the interception is judged to be successful and the task status is updated; if the slave unit sends back "failed interception" or the tracking is lost for more than the set time, the interception is judged to be failed and the interception scheduling is re-executed.

[0557] 3. The principles and process logic of task closure

[0558] Task closure is the summary and archiving of the entire task process. The core logic is "submachine security disposal → task data storage → task effectiveness evaluation", which ensures that submachine resources are reasonably recovered and that task results are traceable and reviewable.

[0559] Security handling logic of the slave unit

[0560] The core of the slave unit disposal is "differentiated processing based on power status", balancing the slave unit recycling rate and safety risks, and avoiding slave unit crashes due to insufficient power.

[0561] Battery level assessment and handling strategy:

[0562] Return-to-home decision logic: The slave unit transmits battery power data back to the ground terminal in real time. When the power level is greater than or equal to the set threshold (ensuring that the slave unit has enough power to return to the take-off and landing point), the ground terminal issues a "return-to-home command". After receiving the command, the slave unit automatically plans the return-to-home path (prioritizing the reverse path of the initial flight path to ensure that the route is familiar and there are no unknown obstacles) and flies to the take-off and landing point (high-speed slave units return to flat ground near the launch pad, and AI slave units return to the mother aircraft or the ground take-off and landing point).

[0563] Forced landing determination logic: When the slave unit's battery level is less than the set threshold, it is determined that "battery is insufficient and cannot return to base", and the ground terminal issues a "forced landing command". After receiving the command, the slave unit immediately initiates the forced landing procedure - first, it reduces its flight altitude, searches for open areas below (excluding dangerous areas such as buildings and water bodies through visual recognition), and selects a safe forced landing point. During the forced landing, it continuously monitors the ground conditions, adjusts its attitude to ensure a smooth landing, and minimizes the risk of damage to the slave unit.

[0564] Submachine recovery and status confirmation:

[0565] Returning aircraft confirmation: After the aircraft arrives at the take-off and landing point and lands successfully, it sends a "Return successful" signal back to the ground terminal; after the ground terminal confirms the signal, it marks the aircraft as "Recovered", completing the aircraft recovery process.

[0566] Forced landing confirmation: If the slave unit can still communicate normally after the forced landing, it will send back a "forced landing successful" signal; if communication is interrupted, the ground terminal will confirm the slave unit's status (whether it is damaged or repairable) through visual monitoring of the mother unit or manual inspection, and update the slave unit's status record.

[0567] Logic for storing and reviewing task data

[0568] The core of task data storage is "structured storage to ensure traceability", and the core of the review report is "quantifying task performance indicators" to provide data support for subsequent task optimization.

[0569] Structured storage of task logs:

[0570] Log content design logic: Logs must contain key data throughout the entire task process to ensure traceability at each stage. The specific content is divided into three categories:

[0571] Command data: includes the issuance time of all commands, command type (takeoff command, interception command, follow-up interception command), and command receiving object (sub-unit ID, mother unit, launcher);

[0572] Status data includes the real-time status of the slave unit (battery level, location, flight attitude, interception status), the monitoring data of the mother unit (target identification results, tracking status), and the environmental data of the launcher (temperature, humidity, equipment status).

[0573] Target data: Includes the initial coordinates, priority, and interception results of all targets (whether they were intercepted, the ID of the intercepting submachine, and the number of times they were intercepted again).

[0574] Storage format and method: Log data is stored in the common JSON format, which has the advantages of high structure, easy parsing and strong cross-platform compatibility; Log data is synchronized in real time to the local storage of the ground terminal and the cloud server to ensure that the data is not lost (local storage is used for quick retrieval, and cloud storage is used for long-term backup).

[0575] The logic behind generating and evaluating a post-mortem report:

[0576] Report Indicator Design: The review report should include quantifiable task performance indicators to intuitively reflect the quality of task completion. Core indicators include:

[0577] Target coverage: (Number of intercepted targets + Number of successfully intercepted targets) / Total number of targets × 100%, reflecting the completeness of target interception;

[0578] Submachine utilization rate: (Number of submachines participating in the task / Total number of available submachines) × 100%, reflecting the utilization efficiency of submachine resources;

[0579] Success rate of interception: Number of targets successfully intercepted / Number of targets that slipped through the net × 100%, reflecting the effectiveness of the interception mechanism.

[0580] Report Generation and Application: The report is generated in Excel format, facilitating data filtering and analysis by operators. The indicator data in the report is linked to the task log; clicking on an indicator displays its detailed data (target coverage 100%).

[0581] (View the interception details for each target); the debriefing report is used to summarize the task. If a certain indicator does not meet expectations (the success rate of re-interception is less than 90%), the cause of the problem can be traced through the logs (submachine scheduling delay, unreasonable path planning), providing a basis for algorithm optimization and process adjustment for subsequent tasks.

[0582] The foregoing has provided a detailed description of a method and system for scheduling a mother-daughter UAV swarm. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for scheduling a cluster of mother and child unmanned aerial vehicles (UAVs), characterized in that, include: A three-tiered collaborative architecture is constructed, consisting of an "airborne mothership - two types of daughter aircraft - ground terminal." The airborne mothership is a heavy-duty rotorcraft capable of carrying daughter aircraft, relay communication, and initial target detection. The two types of daughter aircraft include a low-to-medium speed precision AI racing aircraft adapted for precision detection / interception missions, and a high-speed assault racing aircraft adapted for rapid attack / interception missions. The ground terminal includes a launcher, operating system terminal, interactive terminal, and control software, responsible for target marking, command issuance, and status monitoring. This forms a closed-loop scheduling process of "target detection - command issuance - daughter aircraft execution - status feedback - interception of missed targets": the target detection phase is handled by the airborne mothership. The mother aircraft completes initial target identification and positioning; during the command issuance phase, the ground terminal generates and forwards dispatch commands; during the execution phase, the slave aircraft flies according to the pre-transmitted coordinates and initiates precision guidance at the terminal stage; during the status feedback phase, the slave aircraft transmits flight status and target lock status in real time; when the conditions for missing targets are met, the interception mechanism is triggered, and the interception slave aircraft is preferentially dispatched from the idle racing aircraft. During dispatch, factors such as the target's position, the slave aircraft's position, the target's priority, and the urgency must be considered to allocate high-speed racing aircraft with high-speed attack capabilities or AI racing aircraft with its precise identification capabilities. If there are no idle targets, the nearest AI racing aircraft slave aircraft is dispatched. The delay in issuing the interception command meets the real-time requirements.

2. The method for scheduling a cluster of mother and child unmanned aerial vehicles according to claim 1, characterized in that, The preset multiple ranges from 1.5 to 2.5 and is dynamically adjusted based on the mission scenario; the upper limit of the total number of sub-machines is the sum of the maximum payload of the airborne mothership and the maximum capacity of the launcher. The airborne mothership can carry at least 8 medium- and low-speed precision AI racing drones, and the launcher can accommodate at least 6 high-speed assault racing drones.

3. The method for scheduling a cluster of mother and child unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The target priorities are divided according to the degree of threat or importance: high priority targets are personnel, key equipment or dangerous goods carriers; medium priority targets are ordinary mobile targets; low priority targets are non-threatening static targets or targets in non-core areas; the target priority can be automatically identified by manual labeling or by the AI ​​recognition module of the airborne mothership. When the confidence level of the automatic identification meets the preset threshold, it is directly determined; otherwise, manual review is triggered.

4. The method for scheduling a cluster of mother and child unmanned aerial vehicles according to claim 1, characterized in that, The timing of the dual-link collaborative interaction meets the following requirements: In the pre-transmission phase, the time taken for the airborne mother aircraft and the low-to-medium speed precision AI racing drone to pre-transmit data via the magnetic interface meets the requirements for rapid deployment; the time taken for the launcher and the high-speed assault racing drone to load data via the magnetic interface meets the requirements for launch efficiency; in the execution phase, after the slave unit detaches from the carrier, it automatically switches to the MESH wireless link, and the status feedback frequency meets the monitoring requirements; if the MESH link transmission quality does not meet the requirements, the ground terminal triggers an abnormal alarm after a timeout.

5. The method for scheduling a cluster of mother and child unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The criteria for determining if a target has slipped through the net include at least one of the following: the target has not been tracked by any sub-unit for a period of time exceeding a preset duration; the sub-unit sends back a "task execution failed" signal, with reasons for failure including insufficient power, target loss, or equipment failure; the ground terminal detects a newly added unassigned target; the scheduling priority of the interception sub-unit is determined by factors such as the target's location, the sub-unit's location, the target's priority, and the urgency; and the remaining interception sub-units return to an idle state after the interception task is completed.

6. A parent-child unmanned aerial vehicle (UAV) swarm scheduling system implementing the method of any one of claims 1-5, characterized in that, include: The heavy-duty mother aircraft has preset payload and endurance capabilities and can withstand wind interference of preset levels. It integrates a flight control system, a MESH relay module, and multiple drop units, each equipped with a magnetic interface to support the mounting, data pre-transmission, and precise drop of medium- and low-speed precision AI racing drones. The unlocking sound of the drop units meets the requirements for rapid deployment. The medium- and low-speed precision AI racing drone has target recognition and guidance capabilities, and its flight speed is adapted to the requirements of precision operations. Its single-charge endurance meets the mission execution requirements. It is equipped with an image acquisition module and AI recognition algorithms. It can initiate precise terminal guidance within a preset distance from the target; the fuselage is equipped with a magnetic interface, physically compatible with the magnetic interface of the heavy-duty mothership's launch unit, supporting rapid docking and data transmission; the high-speed assault-type racing drone adopts a streamlined structural design, possessing high-speed flight capability and endurance sufficient for rapid assault needs; equipped with a multi-mode guidance module, it can initiate terminal guidance within a preset distance from the target, and has a magnetic interface at the tail, compatible with the launcher's slot interface; the launcher has at least 6 independent launch slots, each slot integrating a magnetic interface. The system includes a power supply interface and guide rail mechanism; an integrated power management module that supports automatic charging and discharging of the high-speed assault drone, with overcharge and over-discharge protection; an integrated environmental monitoring module that monitors temperature, humidity, and abnormal smoke, automatically cutting off power and triggering an alarm in case of anomalies; safety protection features, with a fire-resistant outer shell that automatically opens during launch and closes after launch, achieving integrated "storage, transportation, and launch"; a ground control terminal including an operating system terminal, an interactive terminal, and control software; the operating system terminal includes a host computer, a MESH repeater, and a gain antenna, supporting long-distance communication and connecting to the launch pad via a wired connection, with data transmission rates meeting command and information interaction requirements; the interactive terminal supports wired and wireless dual-mode connections, with wireless matching time meeting rapid deployment requirements; the control software has equipment management, target operation, command issuance, and data feedback functions, with command generation time meeting real-time requirements; and a collaboration module including a MESH communication module, a resource allocation algorithm module, a dual-link switching module, and a interception algorithm module, with each component achieving data interaction and collaborative work through a preset communication protocol.

7. The mother-daughter UAV swarm scheduling system according to claim 6, characterized in that, The fuselage of the heavy-duty mother aircraft is made of high-strength and lightweight materials. The magnetic interface of the drop unit has dual functions of "two-way data interaction + start signal trigger". During drop, a buffer mechanism is used to avoid collision between the daughter aircraft and the mother aircraft, ensuring the initial flight attitude stability of the daughter aircraft. The MESH relay module supports primary and backup link backup and also has relay forwarding function. It can forward the data returned by the daughter aircraft to the ground terminal, and can also receive commands from the ground terminal and send them to the daughter aircraft.

8. The mother-daughter UAV swarm scheduling system according to claim 6, characterized in that, The AI ​​recognition module of the medium-low speed precision AI racing drone supports the recognition of multiple types of targets such as "people / vehicles" and has the ability to suppress dynamic blur and re-recognize targets that are obscured. The multi-modal guidance module of the high-speed assault racing drone is adapted to the target recognition requirements of high-speed flight scenarios, and has reserved sensor installation positions. It can be expanded to include proximity fuses and warheads.

9. The mother-daughter UAV swarm scheduling system according to claim 6, characterized in that, The launch pad precisely mates with the high-speed racing drone's adapter structure, ensuring that the initial flight direction deviation meets accuracy requirements during vertical launch. The power supply interface supports charging the high-speed racing drone's battery, and the slot is equipped with a status display unit that can show the drone's battery level in real time. The wired connection with the operating system terminal uses an anti-interference cable to avoid abnormal command transmission caused by wireless interference.

10. The mother-daughter UAV swarm scheduling system according to claim 6, characterized in that, The control software includes the following functions: a device management module, which displays the status of all devices in real time, including the location and remaining power of the main unit, the task progress of the sub-units, the occupancy status of the launcher slots, and environmental parameters, and automatically alarms when the device connection is lost; a target operation module, which synchronizes the real-time image of the main unit, supports zooming operations, and automatically triggers the target positioning and image acquisition of the main unit after clicking on the image target, generating a file containing the target identifier, coordinates, and image; a command issuance module, which automatically generates "start command + target coordinates + guidance logic" after selecting the UAV, and the command is forwarded to the main unit or launcher via a repeater, and the sub-unit sends back a confirmation signal after receiving it; and a data feedback module, which displays the flight trajectory and target recognition image of the sub-unit in real time, automatically stores the task log, and supports exporting and reviewing.

Images