Disaster emergency response supercomputing unmanned aerial vehicle and use method thereof

Abstract

The invention discloses a disaster emergency response super-computing unmanned aerial vehicle and a use method thereof. The disaster emergency response super-computing unmanned aerial vehicle comprises an unmanned aerial vehicle body, and wings are fixedly connected to the four corners of the unmanned aerial vehicle body. According to the invention, a supercomputing module and a multi-modal intelligent agent platform are deeply integrated, a dynamic computing power distribution strategy of 20% of data preprocessing, 40% of disaster situation evaluation, 30% of resource scheduling and decision assistance and 10% of redundancy is adopted, and the dynamic weight fusion of a Transform model, CNN / U-Net high-precision disaster situation identification and a resource scheduling algorithm with a'rescue strength fatigue coefficient 'are combined. According to the method, efficient processing and accurate decision output of multi-source disaster situation data are realized, the problems that a traditional unmanned aerial vehicle is insufficient in computing power and insufficient in data fusion in a disaster scene and decision suggestions break away from actual rescue requirements are solved, and disaster situation assessment precision and resource scheduling rationality are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of supercomputing drone technology for disaster emergency response, specifically to a supercomputing drone for disaster emergency response and its usage method. Background Technology

[0002] In emergency response to natural disasters (such as fires, floods, and earthquakes) and public safety incidents, rapidly obtaining disaster information, accurately assessing the scale of the disaster, and efficiently allocating rescue resources are core requirements for improving response efficiency and rescue success rates. Traditional emergency rescue often relies on manual reconnaissance, ordinary drone aerial photography, or fixed monitoring stations to obtain information, but these methods generally have significant limitations: On the one hand, there is insufficient capacity for disaster information collection and processing. Ordinary drones typically carry only a single aerial photography device, capable of acquiring only visible light image data. They cannot simultaneously collect key environmental parameters such as temperature, humidity, harmful gas concentrations, and ground displacement. Furthermore, they lack the ability to integrate unstructured data such as distress messages from social media, resulting in a limited and fragmented understanding of the disaster situation. Simultaneously, traditional drones have limited computing power, making it difficult to perform real-time fusion and in-depth analysis of multi-source data. Data often needs to be transmitted back to the command center for processing, leading to high data transmission latency and delayed disaster assessment. This is particularly problematic in large-scale disaster scenarios, where untimely information can easily cause delays in rescue decisions.

[0003] On the other hand, they suffer from poor adaptability to complex environments and poor communication stability. After a disaster, ground communication infrastructure is often damaged. Ordinary drones rely on a single 4G / 5G communication link, which is prone to data transmission interruption due to signal loss, making it impossible to continuously report the disaster situation to the command center. Some devices with satellite communication capabilities suffer from slow switching response and low data transmission efficiency, making it difficult to meet the real-time requirements of emergency scenarios. In addition, the hardware of traditional emergency drones is mostly integrated, and components such as processors, storage, and heat dissipation are difficult to upgrade or repair individually, resulting in high equipment maintenance costs. Furthermore, under long-term high-load operation, they are prone to performance degradation due to insufficient heat dissipation, making them unsuitable for the continuous operation needs in disaster relief.

[0004] Meanwhile, the capacity for disaster relief resource allocation and decision support is weak. Existing technologies can only provide basic disaster images or data, lacking accurate quantitative assessments of the disaster's scope and severity. Furthermore, resource allocation plans are often generated based on fixed rules, failing to fully consider practical factors such as the fatigue of rescue personnel and the real-time status of equipment. This results in a disconnect between allocation plans and on-site needs, making it difficult to effectively guide rescue operations.

[0005] Therefore, it is necessary to study a supercomputing drone for disaster emergency response and its usage method. Summary of the Invention

[0006] To address the problems mentioned in the background art, the present invention aims to provide a supercomputing drone for disaster emergency response and its usage method, which has the advantages of multi-dimensional disaster perception, efficient data processing, stable communication in complex environments, accurate decision support, and easy hardware maintenance, and solves the problems of easy communication interruption, single perception dimension, and high hardware upgrade cost of traditional emergency drones.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a supercomputing unmanned aerial vehicle (UAV) for disaster emergency response, comprising an UAV body, with wings fixedly connected to each of the four corners of the UAV body, a rotary propeller provided on the top of the UAV body, the rotary propeller being directly driven by a micro-engine, a controller provided inside the UAV body, the controller being electrically connected to the micro-engine, and landing gear symmetrically fixedly connected to both sides of the bottom of the UAV body. The controller includes a supercomputing module, a sensing module, a communication module, and a multimodal intelligent agent platform. The output of the sensing module is connected to the input of the supercomputing module, the output of the supercomputing module is connected to the input of the communication module, the multimodal intelligent agent platform is integrated inside the supercomputing module, and the output of the communication module is connected to an external emergency command center.

[0008] As a preferred embodiment of the present invention, the supercomputing module includes a hardware configuration unit and a computing power allocation unit. The output end of the hardware configuration unit is electrically connected to the input end of the computing power allocation unit, and the output end of the computing power allocation unit is data-connected to the multimodal intelligent agent platform. The hardware configuration unit includes an ARM architecture multi-core processor, a solid-state drive, and a heat dissipation system. The computing power allocation unit allocates computing power according to the proportions of data preprocessing (20%), disaster assessment (40%), resource scheduling and decision support (30%), and redundancy (10%).

[0009] As a preferred embodiment of the present invention, the sensing module includes a high-definition aerial camera unit, a multi-parameter ground sensor unit, and a social media text acquisition unit. The output terminals of the high-definition aerial camera unit, the multi-parameter ground sensor unit, and the social media text acquisition unit are all connected to the input terminal of the supercomputing module.

[0010] As a preferred embodiment of the present invention, the high-definition aerial camera unit is installed on the bottom of the drone body. The high-definition aerial camera unit is equipped with a CMOS sensor and an infrared thermal imaging lens. The multi-parameter ground sensor unit includes a temperature and humidity sensor, a displacement sensor, and a gas sensor. The temperature and humidity sensor and the displacement sensor are installed on the lower side of the drone body via a retractable bracket. The gas sensors are all integrated on the top of the drone body. The social media text acquisition unit accesses mainstream social platforms through an open API and supports the access of emergency command center distress hotline text transcription data.

[0011] In a preferred embodiment of the present invention, the communication module includes a dual-mode communication unit and a data transmission strategy unit. The input end of the dual-mode communication unit is connected to the output end of the supercomputing module, the input end of the data transmission strategy unit is connected to the output end of the dual-mode communication unit, and the output end of the data transmission strategy unit is connected to an external emergency command center. The dual-mode communication unit supports 5G mode and low-orbit satellite communication, and the data transmission strategy unit prioritizes data and supports resuming interrupted transmissions.

[0012] In a preferred embodiment of the present invention, the multimodal intelligent agent platform includes a data fusion unit, a disaster assessment unit, a resource scheduling unit, and a decision support unit. The input end of the data fusion unit is connected to the output end of the sensing module. The input end of the disaster assessment unit is connected to the output end of the data fusion unit. The input end of the resource scheduling unit is connected to the output end of the disaster assessment unit and the external emergency command center resource database, respectively. The input end of the decision support unit is connected to the output end of the resource scheduling unit. The output end of the decision support unit is connected to the input end of the communication module.

[0013] As a preferred embodiment of the present invention, the data fusion unit adopts the Transformer model, the disaster assessment unit adopts the CNN model and the U-Net model, the resource scheduling unit adopts the LSTM model and the integer programming algorithm, and the decision support unit includes a rule base and an inference model to generate a comprehensive situation map and structured decision suggestions.

[0014] As a preferred embodiment of the present invention, the following steps are included: S1. Deployment and Activation: Based on the location of the disaster, the emergency command center transports the UAV to the deployment point, activates the controller through the ground terminal, completes the initialization of the supercomputing module and communication module, establishes a two-way data link with the emergency command center, and at the same time, the landing gear is deployed to the support position; S2. Disaster Sensing: The controller drives the rotary propeller to lift the drone into the air, the sensing module starts working, the high-definition aerial camera unit takes pictures of the disaster area from multiple angles, the multi-parameter ground sensor unit collects temperature, humidity, displacement and gas data in real time, the social media text collection unit filters distress information within the geofence, and all data are synchronously transmitted to the supercomputing module. S3. Data Processing: The computing power allocation unit of the supercomputing module schedules computing power according to a preset ratio, the hardware configuration unit stores and encrypts the perceived data, and the various units of the multimodal intelligent agent platform work together—the data fusion unit fuses multi-source data through the Transformer model, the disaster assessment unit uses the CNN model and the U-Net model to identify the disaster range, the resource scheduling unit generates resource allocation schemes based on the LSTM model and integer programming algorithm, and the decision support unit outputs a comprehensive situation map and decision suggestions in combination with the rule base; S4. Information Interaction: The dual-mode communication unit of the communication module transmits the processing results to the emergency command center according to priority. If the signal is interrupted, the breakpoint resume mechanism is triggered. The feedback instructions from the command center are transmitted back to the supercomputing module through the communication module to dynamically adjust the UAV's flight path or perception parameters. S5. Mission Termination: When the emergency command center issues the termination command, the controller drives the UAV to return to base, the supercomputing module automatically archives all data of this mission and uploads it to the command center database, and then shuts down non-essential modules and enters standby mode.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention deeply integrates a supercomputing module (including an ARM architecture multi-core processor, a high-speed encrypted solid-state drive, and a hybrid heat dissipation system) with a multimodal intelligent agent platform. It adopts a dynamic computing power allocation strategy of "20% data preprocessing, 40% disaster assessment, 30% resource scheduling and decision support, and 10% redundancy". Combined with the dynamic weight fusion of the Transformer model, the high-precision disaster identification of CNN / U-Net, and the resource scheduling algorithm with "rescue force fatigue coefficient", it realizes efficient processing of multi-source disaster data and accurate decision output. It solves the problems of insufficient computing power, insufficient data fusion, and decision suggestions that are out of touch with actual rescue needs of traditional UAVs in disaster scenarios. It significantly improves the accuracy of disaster assessment (pixel level for fires, ≤10 meters for floods) and the rationality of resource scheduling.

[0016] 2. This invention designs a dual-mode communication module of "5G + low-orbit satellite" (automatic switching within 2 seconds when the signal is weak) and a hierarchical data transmission strategy (high-priority data latency ≤10 seconds, medium-priority video dynamically adjusted compression rate), combined with multi-dimensional acquisition by the perception module (high-definition aerial photography + multi-parameter sensing + social media text) and a data preprocessing mechanism (3σ anomaly removal, semantic cleaning + targeted compression). At the same time, it adopts a modular hardware design (supporting individual upgrades and maintenance) and full-process data encryption (AES-256), ensuring the comprehensiveness of data acquisition, the continuity of transmission, and the maintainability of equipment in complex disaster environments. It overcomes the shortcomings of traditional emergency drones, such as easy communication interruption, single perception dimension, and high hardware upgrade costs, and ensures the real-time and stability of emergency response. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a top-view diagram showing the interconnected structure of the present invention; Figure 3 This is a block diagram of the controller of the present invention.

[0018] In the image: 1. Unmanned aerial vehicle (UAV) fuselage; 2. Wings; 3. Rotary propellers; 4. Controller; 5. Landing gear. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figures 1 to 3 As shown, the present invention provides a supercomputing drone for disaster emergency response and its usage method, including a drone body 1, with wings 2 fixedly connected to the four corners of the drone body 1, a rotary propeller 3 set on the top of the drone body 1, and the rotary propeller 3 being directly driven by a micro engine, a controller 4 set inside the drone body 1, and the controller 4 being electrically connected to the micro engine, and landing gear 5 being symmetrically fixedly connected to both sides of the bottom of the drone body 1. The controller 4 includes a supercomputing module, a sensing module, a communication module, and a multimodal intelligent agent platform. The output of the sensing module is connected to the input of the supercomputing module, the output of the supercomputing module is connected to the input of the communication module, the multimodal intelligent agent platform is integrated inside the supercomputing module, and the output of the communication module is connected to the external emergency command center.

[0021] refer to Figure 3The supercomputing module includes a hardware configuration unit and a computing power allocation unit. The output of the hardware configuration unit is electrically connected to the input of the computing power allocation unit, and the output of the computing power allocation unit is connected to the multimodal intelligent agent platform. The hardware configuration unit includes an ARM architecture multi-core processor (such as NVIDIA Jetson AGX Orin, with a computing power of 200 TOPS), a solid-state drive (such as a 2TB SSD with a read / write speed of 2000MB / s and support for AES-256 encryption), and a heat dissipation system (such as a liquid cooling + air cooling hybrid heat dissipation system with a heat dissipation power of 50W and continuous operation for ≥8 hours). The hardware configuration unit adopts a modular design, which allows the ARM architecture multi-core processor, solid-state drive, and heat dissipation system to be detachably connected through a standard interface, facilitating individual upgrades or repairs. The computing power allocation unit allocates computing power according to the proportions of data preprocessing (20%), disaster assessment (40%), resource scheduling and decision support (30%), and redundancy (10%).

[0022] As a technical optimization solution of the present invention, by adopting a modularly designed hardware configuration unit (including an ARM architecture multi-core processor, a high-speed encrypted solid-state drive and a hybrid heat dissipation system), combined with a computing power allocation strategy of "20% for data preprocessing, 40% for disaster assessment, 30% for resource scheduling and decision support, and 10% for redundancy", the system can achieve flexible hardware upgrades and maintenance while ensuring that core disaster handling tasks receive priority computing power support, thus guaranteeing the stability and efficiency of the supercomputing module under continuous high-load operation for 8 hours.

[0023] refer to Figure 3 The perception module includes a high-definition aerial camera unit, a multi-parameter ground sensor unit, a social media text acquisition unit, and a data preprocessing subunit. The outputs of the high-definition aerial camera unit, the multi-parameter ground sensor unit, and the social media text acquisition unit are all connected to the input of the supercomputing module. The data preprocessing subunit performs noise reduction and image stabilization preprocessing on the images acquired by the high-definition aerial camera unit, removes outliers from the raw data of the multi-parameter ground sensor unit (e.g., filtering sudden fluctuations in temperature and humidity sensor data using the 3σ criterion), and performs semantic cleaning on the information from the social media text acquisition unit (e.g., removing duplicate or invalid distress messages). The preprocessed data is compressed using algorithms (e.g., JPEG2000 compression for images and LZ77 compression for sensor data) to reduce transmission volume, thereby reducing the supercomputing module's computing power consumption by 15% to 20% and improving data processing efficiency.

[0024] As a technical optimization of the present invention, by configuring a high-definition aerial camera unit, a multi-parameter ground sensor unit, and a social media text acquisition unit for the perception module, and adding a data preprocessing subunit, coupled with a targeted compression algorithm, the computing power consumption of the supercomputing module is reduced by 15% to 20% while achieving comprehensive acquisition of multi-dimensional disaster data, thereby improving the overall data processing efficiency.

[0025] refer to Figure 3 The high-definition aerial camera unit is mounted on the bottom of the UAV body 1 via a three-axis stabilized gimbal (gimbal rotation angle range: horizontal ±180°, vertical -90° to +30°). The high-definition aerial camera unit is equipped with a CMOS sensor (such as a 20-megapixel CMOS sensor) and an infrared thermal imaging lens (such as an infrared thermal imaging lens with a resolution of 640×512 and a temperature measurement range of -20℃ to 1500℃). The multi-parameter ground sensor unit includes a temperature and humidity sensor (measurement range -40℃ to 85℃, 0% to 100%RH, accuracy ±0.5℃ / ±3%RH). The system includes a displacement sensor (BeiDou / GPS dual-mode positioning, positioning accuracy 1 meter / 3 meters) and a gas sensor (range 0 to 1000ppm, accuracy ±5%FS). The temperature and humidity sensor and the displacement sensor are mounted on the lower side of the UAV body 1 via a retractable bracket. The gas sensors are all integrated into the protruding signal compartment on the top of the UAV body 1 to avoid interference from the heat dissipation of the UAV body 1 on the sensor data and reduce the error of temperature and humidity measurement. The social media text acquisition unit accesses mainstream social platforms through open APIs and supports the access of emergency command center distress hotline text transcription data.

[0026] As a technical optimization of the present invention, by mounting the high-definition aerial camera unit on a three-axis stabilized gimbal that can rotate over a wide range, installing the temperature, humidity and displacement sensors on the lower side via a retractable bracket, integrating the gas sensor into the top protruding signal cabin, and connecting to social platform APIs and distress hotlines to transcribe data, the accuracy (avoiding interference from the aircraft's heat dissipation) and coverage of the data collected by each sensor are ensured, and the real-time aggregation of multi-source disaster information is achieved.

[0027] refer to Figure 3The communication module includes a dual-mode communication unit and a data transmission strategy unit. The input of the dual-mode communication unit is connected to the output of the supercomputing module, and the input of the data transmission strategy unit is connected to the output of the dual-mode communication unit. The output of the data transmission strategy unit is connected to an external emergency command center. The dual-mode communication unit supports 5G mode (download ≥1Gbps, upload ≥100Mbps), low-Earth orbit satellite communication (weight ≤500g, transmission ≥1Mbps), and a signal strength monitoring submodule. The signal strength monitoring submodule monitors the 5G signal coverage quality in real time. When the 5G signal strength is below a threshold, it automatically switches to low-Earth orbit satellite communication mode. The latency is ≤2 seconds. Simultaneously, the data transmission strategy unit employs a dynamic compression algorithm. High-priority data (such as disaster assessment reports) undergoes lightweight compression (10% compression rate) to ensure real-time performance. For medium-priority aerial videos, the compression rate is dynamically adjusted based on bandwidth (30% compression rate when bandwidth ≥ 50Mbps, 60% compression rate when bandwidth < 20Mbps). This ensures data transmission continuity in complex communication environments. The data transmission strategy unit prioritizes data (highest: disaster assessment reports / trapped locations, latency ≤10 seconds; medium: aerial videos, H.265 encoded, latency ≤30 seconds; low: historical data, transmitted after the task), and supports resuming interrupted transmissions.

[0028] As a technical optimization of the present invention, by adopting a communication unit that supports dual-mode switching between 5G and low-orbit satellite, combined with a data hierarchical transmission strategy (lightweight compression of high-priority disaster data, dynamic adjustment of compression rate for medium-priority video, and delayed transmission of low-priority historical data) and a breakpoint resume function, continuous and real-time transmission of high-priority data (delay ≤ 10 seconds) and medium-priority data (delay ≤ 30 seconds) is ensured in complex communication environments.

[0029] refer to Figure 3 The multimodal intelligent agent platform includes a data fusion unit, a disaster assessment unit, a resource scheduling unit, and a decision support unit. The input end of the data fusion unit is connected to the output end of the perception module. The input end of the disaster assessment unit is connected to the output end of the data fusion unit. The input end of the resource scheduling unit is connected to the output end of the disaster assessment unit and the external emergency command center resource database, respectively. The input end of the decision support unit is connected to the output end of the resource scheduling unit. The output end of the decision support unit is connected to the input end of the communication module.

[0030] As a technical optimization of the present invention, by constructing an integrated architecture of "data fusion - disaster assessment - resource scheduling - decision support", the data fusion unit receives data from the sensing module, and after disaster assessment and resource scheduling (linking with the command center resource library), the decision support unit generates output, forming a closed loop from data input to decision output, ensuring the intelligent agent's rapid response to disasters and accurate decision support.

[0031] refer to Figure 3 The data fusion unit employs a Transformer model (input: 2048-dimensional image features, 64-dimensional sensor features, 512-dimensional text features; output: 1024-dimensional fused features). The Transformer model in the data fusion unit incorporates a dynamic attention weight adjustment mechanism, automatically allocating feature weights according to the disaster type. For example, in fire scenarios, the weight of infrared image features is increased (accounting for 40%), while in flood scenarios, the weight of visible light image and displacement sensor features is increased (totaling 50%). The disaster assessment unit uses a CNN model (fire range, pixel-level accuracy) and a U-Net model (flood range, accuracy ≤10). The resource scheduling unit uses an LSTM model (demand forecasting) and an integer programming algorithm (scheme optimization). The resource scheduling unit is equipped with a "rescue force fatigue coefficient" parameter (the rescue force fatigue coefficient is synchronously obtained from the emergency command center resource database to make the scheduling scheme more in line with the actual rescue scenario). The decision support unit includes a rule base and inference model to generate a comprehensive situation map and structured decision suggestions. The comprehensive situation map supports multi-level scaling (from 1:10000 area view to 1:100 building detail view) and uses red and yellow levels for dynamic labeling of high-risk areas (the color depth is updated every 5 minutes) to improve the visual recognition efficiency of the command personnel.

[0032] As a technical optimization scheme of the present invention, by adopting a dynamic attention weight Transformer model (assigning feature weights according to disaster type) in the data fusion unit, a high-precision CNN / U-Net model in the disaster assessment unit, introducing the "fatigue coefficient of rescue force" and LSTM+integer programming algorithm in the resource scheduling unit, and generating a scalable dynamic labeled situation map in the decision support unit, the accuracy of multimodal data fusion, the accuracy of disaster assessment, and the practicality of decision recommendations are significantly improved.

[0033] The working principle and usage process of this invention: Based on the location of the disaster, the emergency command center transports the UAV 1 to a deployment point that meets the following conditions: unobstructed view (no obstacles higher than 10 meters within a 100-meter radius), ground load-bearing capacity ≥ 50 kg, and avoidance of strong electromagnetic interference sources. The center then logs into the control interface via a ground terminal using a dedicated encryption key, activating the controller 4. The system automatically detects the status of each module of the UAV 1 (battery power ≥ 80%, sensor calibration values ​​within the error range, communication link signal strength ≥ -85dBm), and completes the initialization of the supercomputing module (including hardware configuration unit and computing power allocation unit) and communication module (including dual-mode communication unit and data transmission strategy unit) within the controller 4. During this process, the supercomputing module performs a peak computing power test. (Performance loss ≤5%), the communication module establishes a two-way data link with the emergency command center and synchronizes basic geographic information data such as 1:5000 scale topographic maps and coordinates of key protected targets. At the same time, the landing gear 5 on both sides of the bottom of the UAV body 1 extends to the support state and completes two reciprocating motion tests of the telescopic mechanism. After initialization, the controller 4 drives the rotary propeller 3 on the top of the UAV body 1 (first run it unloaded for 3 seconds to test) to lift the UAV body 1 into the air. It adopts a step-climb mode (hovering for 3 seconds every 50 meters to adjust) to reach the preset cruising altitude (300-500 meters for fires, 100-200 meters for floods). Then, the sensing module of the controller 4 (including a high-definition aerial camera unit, a multi-parameter ground sensor unit, and social media text) The data acquisition unit and data preprocessing subunit are activated. The high-definition aerial camera unit (mounted on the bottom of the UAV body 1 via a three-axis stabilized gimbal, with a horizontal rotation angle of ±180° and a vertical rotation angle of -90° to +30°) selects either visible light mode (30fps) or infrared thermal imaging mode (15fps, automatically switching between nighttime and dense smoke environments, with dual-mode comparison and calibration every 10 minutes) to capture multi-angle images of the disaster area based on lighting conditions. The multi-parameter ground sensor unit, with its temperature and humidity sensor (measurement range -40℃ to 85℃, 0% to 100%RH, accuracy ±0.5℃ / ±3%RH) mounted on the lower side of the UAV body 1 via a retractable bracket, acquires data at a frequency of 1 time per second. The displacement sensor (BeiDou / GPS) also collects data. The dual-mode system (positioning accuracy 1 meter / 3 meters) collects data at a frequency of 10 times / second. The gas sensor (range 0 to 1000ppm, accuracy ±5%FS) integrated in the protruding signal cabin on the top of the UAV body continuously monitors and caches data once every 5 seconds. The social media text acquisition unit accesses mainstream social platforms through open APIs and supports access to distress hotline text transcription data. It filters distress information within the polygonal geofence remotely demarcated by the emergency command center. All collected data is processed by the data preprocessing subunit of the perception module (image noise reduction and anti-shake, sensor data 3σ anomaly removal, text semantic cleaning) and then synchronously transmitted to the supercomputing module according to the priority of "real-time video stream (high) > sensor abnormal data (medium) > normal status data (low)".The supercomputing module's computing power allocation unit initially allocates computing power according to a preset ratio of "20% for data preprocessing, 40% for disaster assessment, 30% for resource scheduling and decision support, and 10% for redundancy." If a new distress signal is detected, the allocation is temporarily adjusted to 40% for resource scheduling and decision support and 5% for redundancy (this adjustment is restored after 3 minutes). The hardware configuration unit (including an ARM architecture multi-core processor, a 2TB SSD solid-state drive, a liquid-cooled + air-cooled hybrid heat dissipation system, and a modular design with detachable connections) performs local encrypted storage (AES-256 encryption) and real-time incremental uploads of the perceived data. Simultaneously, the multimodal intelligent agent platform integrated within the supercomputing module (including a data fusion unit, disaster assessment unit, resource scheduling unit, and decision support unit) coordinates... The data fusion unit employs a Transformer model (inputting 2048-dimensional image features, 64-dimensional sensor features, and 512-dimensional text features, outputting 1024-dimensional fused features) and dynamically adjusts feature weights according to the disaster type (e.g., increasing the weight of infrared images to 40% for fires) to fuse multi-source data. The disaster assessment unit utilizes a CNN model (fire extent, pixel-level accuracy) and a U-Net model (flood extent, accuracy ≤10 meters). The initial assessment uses a fast mode (85% accuracy, ≤10 seconds), followed by a fine mode (95% accuracy, ≤30 seconds) to identify the disaster extent. The resource scheduling unit synchronizes with the emergency command center's resource database (location of rescue forces, equipment, etc.) every 15 minutes. Based on the status and personnel fatigue level, a resource allocation plan is generated using an LSTM model and integer programming algorithm, incorporating a "rescue force fatigue coefficient". The decision support unit, combined with a rule base, outputs a comprehensive situation map (supporting scaling from 1:10000 to 1:100, with red and yellow levels for high-risk areas updated every 5 minutes) and decision suggestions. During processing, the raw data employs a storage strategy of local encrypted storage + real-time incremental upload, dual copies of processing results (local SSD + command center cloud), and saving one keyframe image every 30 seconds. After processing, the dual-mode communication unit of the communication module (supporting 5G and low-orbit satellite communication, weighing ≤500g) prioritizes the processing results (highest: disaster). Situation assessment report / stuck location, delay ≤10 seconds; Medium: Aerial video, H.265 encoding, delay ≤30 seconds; Low: Historical data, transmitted after the mission) are transmitted through the data transmission strategy unit, using end-to-end AES-256 encryption and updating the session key every 5 minutes. The data transmission strategy unit lightly compresses high-priority data (compression rate 10%) and dynamically adjusts the compression rate of medium-priority aerial video according to bandwidth (30% when bandwidth ≥50Mbps, 60% when <20Mbps). At the same time, the signal strength monitoring submodule of the communication module collects data every second. When the 5G signal is below -105dBm for 3 consecutive seconds, satellite communication switching is initiated (switching time ≤2 seconds). If the signal is interrupted, the breakpoint resume mechanism is triggered.After receiving the data, the emergency command center sends back instructions. The communication module transmits the instructions back to the supercomputing module, which completes a feasibility assessment within 2 seconds and returns "execution confirmation" or "adjustment suggestion." It then drives the UAV body 1 to adjust its flight path (avoiding disaster core areas with temperatures >80℃ or gas concentrations >500ppm, maintaining a safe distance of ≥50 meters from rescue personnel, and reserving an emergency avoidance passage) or sensor module parameters. When the emergency command center issues a termination instruction, the controller 4 drives the UAV body 1 to return via the same route as its takeoff (for easier data comparison). After returning, the supercomputing module automatically archives the original sensor data, processing and analysis reports, flight trajectory records, equipment operation logs, and abnormal event records for this mission. It generates encrypted data packets and transmits them to the emergency command center database via a dedicated line, simultaneously generating a verification code for data integrity verification. After archiving, the controller 4 shuts down unnecessary modules, retaining only the communication module and the core sensors of the sensor module for low-power operation (endurance ≥72 hours, supports remote wake-up). Simultaneously, the controller 4's supercomputing module automatically generates equipment wear reports (such as temperature and humidity sensor and displacement sensor calibration deviations, and battery capacity decay rates), completing the entire usage process. ;

[0034] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A supercomputing unmanned aerial vehicle (UAV) for disaster emergency response, comprising an UAV body (1), characterized in that: The drone body (1) has wings (2) fixedly connected to each of its four corners. The top of the drone body (1) is provided with a rotary propeller (3), which is directly driven by a micro engine. The drone body (1) is provided with a controller (4), which is electrically connected to the micro engine. The bottom sides of the drone body (1) are symmetrically fixedly connected with landing gear (5). The controller (4) includes a supercomputing module, a sensing module, a communication module, and a multimodal intelligent agent platform. The output of the sensing module is connected to the input of the supercomputing module, the output of the supercomputing module is connected to the input of the communication module, the multimodal intelligent agent platform is integrated inside the supercomputing module, and the output of the communication module is connected to an external emergency command center.

2. The supercomputing drone for disaster emergency response according to claim 1, characterized in that: The supercomputing module includes a hardware configuration unit and a computing power allocation unit. The output of the hardware configuration unit is electrically connected to the input of the computing power allocation unit. The output of the computing power allocation unit is data-connected to the multimodal intelligent agent platform. The hardware configuration unit includes an ARM architecture multi-core processor, a solid-state drive, and a heat dissipation system. The computing power allocation unit allocates computing power according to the proportions of data preprocessing (20%), disaster assessment (40%), resource scheduling and decision support (30%), and redundancy (10%).

3. The supercomputing drone for disaster emergency response according to claim 2, characterized in that: The sensing module includes a high-definition aerial camera unit, a multi-parameter ground sensor unit, and a social media text acquisition unit. The output terminals of the high-definition aerial camera unit, the multi-parameter ground sensor unit, and the social media text acquisition unit are all connected to the input terminal of the supercomputing module.

4. The supercomputing drone for disaster emergency response according to claim 3, characterized in that: The high-definition aerial camera unit is installed at the bottom of the UAV body (1). The high-definition aerial camera unit is equipped with a CMOS sensor and an infrared thermal imaging lens. The multi-parameter ground sensor unit includes a temperature and humidity sensor, a displacement sensor and a gas sensor. The temperature and humidity sensor and the displacement sensor are installed on the lower side of the UAV body (1) through a retractable bracket. The gas sensors are all integrated on the top of the UAV body (1). The social media text acquisition unit accesses mainstream social platforms through open APIs and supports the access of emergency command center distress hotline text transcription data.

5. The supercomputing drone for disaster emergency response according to claim 4, characterized in that: The communication module includes a dual-mode communication unit and a data transmission strategy unit. The input of the dual-mode communication unit is connected to the output of the supercomputing module, and the input of the data transmission strategy unit is connected to the output of the dual-mode communication unit. The output of the data transmission strategy unit is connected to an external emergency command center. The dual-mode communication unit supports 5G mode and low-orbit satellite communication, and the data transmission strategy unit prioritizes data and supports resuming interrupted transmissions.

6. The supercomputing drone for disaster emergency response according to claim 5, characterized in that: The multimodal intelligent agent platform includes a data fusion unit, a disaster assessment unit, a resource scheduling unit, and a decision support unit. The input end of the data fusion unit is connected to the output end of the sensing module. The input end of the disaster assessment unit is connected to the output end of the data fusion unit. The input end of the resource scheduling unit is connected to the output end of the disaster assessment unit and the external emergency command center resource database, respectively. The input end of the decision support unit is connected to the output end of the resource scheduling unit. The output end of the decision support unit is connected to the input end of the communication module.

7. The supercomputing drone for disaster emergency response according to claim 6, characterized in that: The data fusion unit uses the Transformer model, the disaster assessment unit uses the CNN model and the U-Net model, the resource scheduling unit uses the LSTM model and the integer programming algorithm, and the decision support unit includes a rule base and an inference model to generate a comprehensive situation map and structured decision suggestions.

8. The method of using a supercomputing drone for disaster emergency response according to claim 7, comprising the following steps: S1. Deployment Start-up: The emergency command center transports the UAV body (1) to the deployment point according to the location of the disaster, activates the controller (4) through the ground terminal, completes the initialization of the supercomputing module and communication module, establishes a two-way data link with the emergency command center, and at the same time the landing gear (5) is deployed to the support state; S2. Disaster perception: The controller (4) drives the rotary propeller (3) to take off, the perception module starts working, the high-definition aerial camera unit takes pictures of the disaster area from multiple angles, the multi-parameter ground sensor unit collects temperature, humidity, displacement and gas data in real time, the social media text collection unit filters the distress information within the geofence, and all data are transmitted synchronously to the supercomputing module. S3. Data Processing: The computing power allocation unit of the supercomputing module first allocates computing power according to a preset ratio, and the hardware configuration unit then stores and encrypts the sensed data; at the same time, the various units of the multimodal intelligent agent platform work together, namely, the data fusion unit fuses multi-source data through the Transformer model, the disaster assessment unit uses the CNN model and the U-Net model to identify the disaster range, the resource scheduling unit generates resource allocation schemes based on the LSTM model and integer programming algorithm, and the decision support unit outputs a comprehensive situation map and decision suggestions in combination with the rule base; S4. Information Interaction: The dual-mode communication unit of the communication module transmits the processing results to the emergency command center according to priority. If the signal is interrupted, the breakpoint resume mechanism is triggered. The feedback instructions from the command center are transmitted back to the supercomputing module through the communication module to dynamically adjust the UAV's flight path or perception parameters. S5. Mission Termination: When the emergency command center issues the termination command, the controller (4) drives the UAV to return home. The supercomputing module automatically archives all data of this mission and uploads it to the command center database. Then, it shuts down unnecessary modules and enters standby mode.

Images