Modularized orchard air-assisted spraying system and quick-change device

Abstract

The invention relates to the field of agricultural machinery, and discloses a modular orchard air-assisted spraying system and a quick-change device aiming at the problems that traditional orchard spraying equipment is single in function, complex in part replacement, lack of effective monitoring and the like. The system comprises a cloud platform, a modular spraying system and an equipment monitoring and regulating system, the cloud platform provides data storage, remote access and decision support services, and the modular spraying system comprises a bearing module, a power module, a fan module, a liquid medicine conveying module, a spraying module and a quick change device. The equipment monitoring regulation and control system comprises a sensor, an edge calculation unit and a local control execution unit. The orchard spraying robot has the advantages of modular design, efficient and quick change, intelligent monitoring and the like, and the efficiency and adaptability of orchard spraying operation can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural machinery, and in particular to a modular orchard air-conveying spray system and a quick-change device. Background Art

[0002] During orchard cultivation, pest and disease control is one of the key links in ensuring fruit quality and yield. Traditional orchard air-assisted spray equipment has gradually exposed a series of problems that need to be solved in the long-term application process. First, traditional orchard spray equipment is mostly manual hand-held sprayers or large fixed sprayers. However, manual hand-held sprayers have problems such as high labor intensity, low spray efficiency, and uneven spraying; large fixed sprayers have shortcomings such as poor flexibility, difficulty in adapting to different orchard terrains and fruit tree heights, high equipment costs, and complex maintenance. Secondly, existing orchard spray equipment has obvious defects in the integration and scalability of functional units, making it difficult to flexibly configure according to different orchard sizes, fruit tree varieties, and types of pests and diseases.

[0003] In addition, the difficulty of equipment maintenance and component replacement is another major pain point of traditional spray equipment. During orchard spraying operations, components such as spray nozzles and fan blades are prone to wear, blockage, or damage and need to be replaced. Traditional equipment usually adopts an integrated structure, and the connections between the various components are complex and tight. The disassembly and installation process is cumbersome and requires professional technicians to spend a lot of time and energy. This not only increases the maintenance cost of the equipment, but also causes excessive equipment downtime, affecting the normal production progress of the orchard. At the same time, traditional spray equipment lacks effective system control, adjustment, and monitoring or testing devices. It is impossible to achieve precise spraying, real-time monitoring of equipment operating status, and scientific evaluation of spraying effects, resulting in serious waste of pesticides, increased environmental pollution, and unstable pest control effects.

[0004] Although some orchard spray equipment has attempted modular designs, they still lack the intelligent coordination of modules and precise adaptability to the complex orchard environment. Therefore, the development of a modular orchard air-assisted spray system and quick-change device is of great practical significance for promoting the modernization and intelligentization of orchard plant protection technology. Summary of the Invention

[0005] The present invention aims to solve the above-mentioned problems. To this end, the present invention adopts the following technical solutions: a modular orchard air-assisted spray system and a quick-change device, the method comprising: A modular orchard air-assisted spray system and quick-change device, characterized by comprising: a cloud platform, a modular spray system, and an equipment monitoring and control system; The cloud platform provides data storage, remote access and decision support services; The modular spray system comprises: The load-bearing module provides structural support for the system and is used to fix and connect other modules; the power module provides power output for the system; the fan module includes the fan and air duct; the liquid medicine delivery module includes the liquid medicine storage tank and delivery pipeline; the spray module includes the nozzle assembly; the quick-change device includes the connector and quick-locking mechanism; The equipment monitoring and control system includes a sensor, an edge computing unit, and a local control execution unit; The sensor is connected to the edge computing unit via a communication interface, and the edge computing unit is connected to the cloud platform via wireless communication.

[0006] Furthermore, the cloud platform uses a deep learning algorithm to analyze orchard environmental data and equipment operation data, and the deep learning algorithm includes a target detection algorithm and a parameter optimization algorithm.

[0007] Furthermore, the carrier module includes a main frame structure and a module installation interface. The main frame structure is made of lightweight materials, and the module installation interface is provided with a standardized connection interface for connecting and separating with the power module, fan module, liquid medicine delivery module, and spray module through a quick-change device.

[0008] Furthermore, the quick locking mechanism of the quick-change device adopts a snap-on connection method to achieve tool-free disassembly and assembly between modules.

[0009] Furthermore, the sensors include a GPS locator, a fuel level sensor, a fan speed sensor, a temperature sensor, a flow sensor, a liquid medicine level sensor, an environmental monitoring sensor and an image recognition camera.

[0010] Furthermore, the edge computing unit receives the operating parameters of each module collected by the sensor in real time, preprocesses, formats and caches the operating parameters, uploads the processed operating parameters to the cloud platform, and at the same time receives the control instructions issued by the cloud platform and forwards them to the local control execution unit.

[0011] Furthermore, the local control execution unit includes a motor driver, a solenoid valve controller and a servo controller, which are used to execute cloud platform control instructions forwarded by the edge computing unit; when communication is interrupted, the edge computing unit performs local emergency control based on a preset security policy.

[0012] Furthermore, the target detection algorithm is based on the YOLOv5 model to identify the location, size and density of fruit trees; The total loss function of the YOLOv5 model comprehensively considers bounding box positioning, target confidence and category classification. Defined as: in, 、 、 are weight coefficients, respectively, used to balance the losses of each part; Represents the positioning loss, and the specific formula is: in, represents the intersection-over-union ratio of the predicted box and the true box, , ) represents the Euclidean distance between the center point of the predicted box and the real box, Represents the diagonal distance between the minimum bounding rectangle of the predicted box and the real box, represents the weight coefficient, represents the aspect ratio consistency factor; Represents the confidence loss, the specific formula is: in, and represent the true and predicted confidence respectively; Represents the classification loss, and the specific formula is: in, and are the true and predicted class probabilities; The parameter optimization algorithm adopts the PPO reinforcement learning algorithm, and the steps are as follows: First, define the state space and motion space; The state dimension of the state space is expressed as: , in, Indicates fruit tree characteristic parameters, environmental parameters or equipment status parameters; The action dimension of the motion space is expressed as: , in, For fan speed adjustment; To regulate the flow of liquid medicine; Adjust the nozzle angle; Next, design the reward function. The reward function is designed as follows: in, 、 、 、 is the weight coefficient, To reward work efficiency, Reward for liquid medicine utilization rate, Reward for equipment safety, rewards for energy consumption optimization; Then, design the loss function of the PPO reinforcement learning algorithm , the formula is: in: represents the expectation for all time steps t; A new strategy for PPO reinforcement learning The probability ratio of taking an action at time step t with the old policy; is the advantage function estimate, which represents the difference between the action value function and the state value function in the PPO reinforcement learning algorithm; is the cropping parameter; For the trimming operation, Restricted to [ ] interval; Finally, the fruit tree information identified by the target detection algorithm is directly used as the fruit tree feature parameters of the parameter optimization algorithm, and the optimal nozzle angle, fan speed and liquid flow rate parameters in the current scene are finally output.

[0013] Furthermore, the system adopts a three-level collaborative control architecture of cloud, edge and end: the cloud platform performs global decision optimization based on deep learning algorithms and issues control instructions, the edge computing unit is responsible for data processing, instruction forwarding and emergency local control during communication interruptions, and the local control execution unit is responsible for accurately executing control instructions and status feedback.

[0014] The present invention provides a modular orchard air-assisted spray system and quick-change device. It has the following beneficial effects: 1. Efficient Operation: Snap-on or threaded connections enable quick replacement of core components such as fans and nozzles. Operators can complete component replacement in a short time without complex tools or specialized skills, allowing the equipment to quickly return to working order and effectively reducing downtime.

[0015] 2. Intelligent Control: Multiple sensors monitor the system's operating status in real time and upload the data to a data processing unit on the cloud platform. This data processing unit uses deep learning algorithms for intelligent analysis. The target detection algorithm identifies the location, size, and density of fruit trees and automatically adjusts the nozzle angle, fan speed, and spray flow rate based on the results, achieving adaptive optimization control.

[0016] 3. Energy saving and environmental protection: The system uses an intelligent control module to dynamically adjust operating parameters such as power output, spray flow rate, and fan speed based on the growth status of the fruit trees and environmental parameters. The parameter optimization function of the deep learning algorithm ensures optimal spraying results under different environmental conditions. Precision variable spray technology significantly improves spray utilization, reduces spray waste, and reduces environmental pollution. Furthermore, by real-time monitoring of parameters such as engine speed, ambient temperature and humidity, energy consumption is optimized, achieving significant energy savings compared to traditional fixed-parameter operation modes.

[0017] 4. Cloud Platform Collaborative Management: Wireless communication enables real-time connection between sensors and the cloud platform. The cloud platform's data storage and intelligent analysis capabilities support centralized management of multiple devices, historical data analysis, and predictive maintenance. The operation progress monitoring module provides real-time location tracking and operation efficiency analysis, while the equipment operation monitoring module implements fault detection, diagnosis, and abnormal alarm processing, significantly improving the standardization and intelligentization of orchard spraying operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 This is a system architecture diagram of the modular orchard air-transported spray system and quick-change device of the present invention; Figure 2 This is a system architecture diagram of the installation of various modules of the spray system of the present invention; Figure 3 It is the intelligent control execution flow chart of the present invention. DETAILED DESCRIPTION

[0020] To achieve the above objectives, the present invention is implemented through the following technical solutions. The present invention provides a modular orchard air-conveying spray system and a quick-changing device. The system architecture diagram of the modular orchard air-conveying spray system and the quick-changing device is as follows: Figure 1 As shown, the method includes: 1. Spray system module selection The load-bearing module provides structural support for the system. The main frame structure is made of lightweight materials and is equipped with standardized connection interfaces for fixing and connecting other modules. The power module provides power output for the system, including but not limited to electric motors or internal combustion engines, and can be quickly connected and disassembled with the carrier module through a quick-change device The fan module includes a fan and an air duct. The fan air volume is adjustable. The air duct is made of lightweight aluminum alloy and has an adjustable length. The liquid medicine delivery module includes a liquid medicine storage tank and a delivery pipeline. The liquid medicine storage tank is made of corrosion-resistant materials, and the delivery pipeline is equipped with a flow control valve; The spray module, including the nozzle assembly, can be quickly replaced through a quick-change device according to the orchard planting density and tree shape characteristics; The quick-change device includes a connecting seat and a quick locking mechanism. The connecting seat is used to connect the load-bearing module, power module, fan module and liquid delivery module. The quick locking mechanism enables rapid disassembly and assembly between modules. The system architecture diagram of the installation of each module of the spray system is as follows Figure 2 As shown; 2. Construction of equipment monitoring and control system Sensor deployment: GPS positioning sensor: installed in the center of the top of the main frame, used to provide accurate location information, support path planning and operation track recording; Fuel level sensor: installed at the bottom of the fuel tank to monitor the remaining fuel and warn of insufficient fuel; Fan speed sensor: installed on the fan bearing seat, used to monitor the actual fan speed and calculate the air volume output; Liquid level sensor: Installed in the center of the top of the storage tank of the liquid medicine delivery module, used to monitor the remaining liquid amount and prevent empty tank operation Liquid medicine flow sensor: installed in the straight pipe section of the liquid medicine delivery module to accurately measure the liquid medicine flow rate; Nozzle angle sensor: installed in the nozzle adjustment mechanism to monitor the current angle of the nozzle and achieve precise adjustment; Environmental monitoring sensors: including temperature sensors, wind speed sensors and wind direction sensors, installed inside the orchard to detect the temperature, wind speed and wind direction inside the orchard in real time; Main camera: installed in front of the device, used to identify the location, size, and density of fruit trees in front; Edge computing unit configuration: The edge computing unit uses an embedded ARM processor, integrates Wi-Fi and 4G communication modules, supports communication protocols such as MQTT and HTTP, and has data processing, format conversion, timestamp synchronization and temporary caching. It receives control instructions issued by the cloud platform, forwards them to the local control execution unit after rationality check, and executes local emergency control based on preset security policies when communication interruption or emergency situation is detected, monitors the execution status of the local control execution unit, and promptly feedbacks abnormal situations.

[0021] Local control execution unit build: The local control execution unit includes a motor driver, solenoid valve controller, and servo controller. The motor driver controls the fan speed and supports variable frequency speed regulation; the solenoid valve controller regulates the liquid flow; and the servo controller adjusts the nozzle angle. Each controller is equipped with a state feedback interface that reports execution status and parameter values to the edge computing unit in real time, forming a closed-loop control system.

[0022] 3. Equipment assembly Use the quick-change device to assemble the selected carrier module, power module, fan module, liquid medicine delivery module, and spray module together to ensure that the installation position of the interface is accurate and the connection is firm; install the locking mechanism so that the locking mechanism can reliably lock the quick-change interface to prevent the interface from loosening during the spraying process.

[0023] 4. Cloud platform deployment The cloud platform adopts a microservice architecture design, which includes three core service modules: data storage service, remote access service, and decision support service.

[0024] Data storage service: Use MySQL relational database to store device information, operation records, and user data; Remote access service: Real-time data push is achieved based on the WebSocket protocol, supporting multiple terminals online at the same time. The monitoring interface is developed using the Vue.js framework to achieve a visual large-screen design, displaying the device location, operating status, environmental parameters and alarm information in real time; Decision support services: Integrates the deep learning algorithm framework TensorFlow, deploys the target detection algorithm YOLOv5 and parameter optimization algorithm. Supports fruit tree identification, operation path planning, and automatic parameter optimization; The target detection algorithm based on YOLOv5 is as follows: Step 1. Loss function design The total loss function of YOLOv5 takes into account bounding box positioning, target confidence and category classification, and is defined as: in, 、 、 are weight coefficients, which are used to balance the losses of each part.

[0025] Positioning loss use The loss is calculated as: in, represents the intersection-over-union ratio of the predicted box and the true box, , ) represents the Euclidean distance between the center point of the predicted box and the real box, Represents the diagonal distance between the minimum enclosing rectangle of the predicted box and the real box, represents the weight coefficient, Represents the aspect ratio consistency factor, which is used to measure the difference in aspect ratio between the predicted box and the real box. , Represents the width and height of the real box, w, h represent the width and height of the predicted box; In the orchard scene, The loss ensures that the model accurately detects fruit locations and remains robust even under canopy occlusion or lighting changes; Confidence loss and classification loss Both use binary cross entropy (BCE): in, and denote the true and predicted confidence respectively, and are the true and predicted class probabilities; The loss function guides the model to distinguish categories such as fruit tree location, size, and density, and optimizes the detection accuracy of orchard targets.

[0026] Step 2: Pre-training and fine-tuning YOLOv5 is pre-trained on the COCO dataset (about 118k images, 80 object categories) to learn general object detection features; For orchard scenes, we used approximately 1,000 labeled images of fruit tree growth stages (e.g., flowering and fruiting) for fine-tuning. Freeze the first few layers of the backbone network during training and only update the subsequent layers to preserve common features; Data augmentation technology (random scaling) improves the model's adaptability to complex orchard scenarios; Step 3: Lightweight and Accelerate To meet the real-time performance requirements of orchard equipment, we performed lightweight pruning on YOLOv5, removing redundant convolutional branches from PANet. The optimized model was deployed on a cloud platform to meet the real-time monitoring needs of orchards.

[0027] The parameter optimization algorithm uses the PPO reinforcement learning algorithm. The steps are as follows: Step 301: State space definition Status dimension composition: Among them, the characteristic parameters of fruit trees are: - (12 dimensions), including parameters such as the location, size, and density of fruit trees; environmental parameters: - (8 dimensions), including wind speed, temperature, wind direction and other parameters; equipment status parameters: - (12 dimensions), including parameters such as fan speed, liquid flow rate, nozzle angle, etc. State standardization: ; in is the state value after normalization of the i-th dimension; is the original state value of the i-th dimension; is the mean of the i-th dimension state; is the standard deviation of the state in the i-th dimension; Normalized state vector: , in ; Step 302: Action Space Design Action dimension definition: in, Adjust the fan speed ; For liquid flow regulation ; Adjust the nozzle angle ; Action constraint function , The constraint principle is: Restricted to [ ] interval, if Less than , just take ; greater than , just take , which plays the role of constraining the update amplitude of the action; Step 303: Reward function design Comprehensive reward function: in 、 、 、 is the weight coefficient; Work efficiency rewards: in, : Covered operating area; : total operation time; : delay time; 、 is the weight coefficient; Rewards for drug utilization rate: in, : Effective amount of liquid medicine; : Total liquid consumption; : target concentration; : actual concentration; 、 is the weight coefficient; Equipment Safety Rewards: in, : engine temperature; : Maximum safe temperature = 110; : Exception indicator (0 or 1); 、 is the weight coefficient; Energy consumption optimization rewards: in, : total power consumption (kW); : Standard power consumption = 25 kW; is a custom coefficient; Step 304, PPO loss function calculation formula in: represents the expectation for all time steps t; is the probability ratio, that is, the new strategy (parameter ) and the old strategy (strategy before updating, recorded as ) The probability ratio of taking an action at time step t is calculated as: For new strategies In state Next select action probability; For the old strategy In state Next select action probability; is the estimated value of the advantage function, and the calculation formula is: Action-value function, which represents the choice of action After that, the expectation of future cumulative rewards; State value function, representing the current state Next, select any action After that, the expectation of future cumulative rewards; is the cropping parameter; For the trimming operation, Restricted to [ ] interval, if Less than , just take ; greater than , just take , which plays a role in constraining the update range of the new strategy.

[0028] Step 305: Parameter update Based on the calculated loss function, the parameters of the policy network and value network are updated using the stochastic gradient descent method; Step 306: Training The current strategy interacts continuously with the environment, performing multiple rounds of sampling, collecting a complete sequence of states, actions, rewards, and termination signals in each round. Strictly adhere to the action constraint function to ensure that action parameters such as fan speed, liquid flow, and nozzle angle are within a reasonable range. If a device state parameter (such as engine temperature approaching the maximum safe temperature) triggers a safety risk threshold, the action is immediately forced to adjust and corresponding penalties and rewards are given to ensure safe equipment operation. The fruit tree information identified by the target detection algorithm is directly used as the fruit tree feature parameters of the parameter optimization algorithm. Through multiple rounds of sampling, under the premise of following the action constraint function and equipment safety threshold, the strategy network and value network parameters are iteratively optimized, and finally the optimal nozzle angle, fan speed and liquid flow rate parameters in the current scenario are output.

[0029] 6. Intelligent control process The intelligent control execution process adopts a three-level collaborative control architecture of cloud, edge and end, which includes four stages. Figure 3 As shown: The first is the real-time data fusion stage. The sensor system collects data at different frequencies. The edge computing unit standardizes the format and synchronizes the timestamps of all collected data. After processing, it is uploaded to the cloud platform through the MQTT protocol. Under normal conditions, batch uploads are made every 30 seconds, and in emergency conditions, real-time push is pushed.

[0030] The second stage is parameter calculation and optimization. After receiving the data, the cloud platform performs quality inspection and outlier filtering. Abnormal data triggers the alarm mechanism, and valid data is input into the target detection algorithm and parameter optimization algorithm for analysis and processing. The algorithm comprehensively considers factors such as fruit tree characteristics, environmental conditions, and equipment status, and generates control instructions containing parameters such as fan speed, liquid flow, and nozzle angle.

[0031] Then comes the control instruction issuance and execution stage. The cloud platform issues control instructions through the wireless communication network. The edge computing unit performs a security check after receiving them. After confirming that they are correct, it forwards the instructions to the local control execution unit. Each controller performs specific parameter adjustment actions and provides real-time feedback on the execution status.

[0032] The last stage is the feedback correction and optimization stage. The sensor continuously monitors the execution effect, and the edge computing unit calculates the error between the actual value and the target value. When it exceeds the allowable range, the parameters are automatically corrected. At the same time, the control effect data is uploaded to the cloud platform for continuous learning and optimization of the algorithm.

[0033] When the edge computing unit detects an emergency situation such as network interruption, equipment abnormality or sudden environmental change, the system automatically switches to local emergency control mode and performs autonomous protective control based on preset safety policies to ensure equipment safety and operation continuity.

[0034] The innovation of this embodiment lies in: (1) Modular quick-change technology: The present invention adopts a modular design, dividing the spray system into multiple independent functional units, including a load-bearing module, a power module, a blower module, a liquid medicine delivery module, a spray module, a quick-change device, and a control module. This design allows users to flexibly select and combine different modules according to the actual needs of the orchard, such as the size of the orchard, the variety of fruit trees, and the type of pests and diseases. The quick-change device can be replaced in seconds through a snap-on or threaded connection, solving the problem of tedious disassembly and assembly of traditional equipment.

[0035] (2) Intelligent Monitoring and Fault Warning: This invention integrates a multi-sensor fusion monitoring system, including a GPS locator, a fuel level sensor, a fan speed sensor, a temperature sensor, a flow sensor, a liquid medicine level sensor, an environmental monitoring sensor, and an image recognition camera. The equipment operation monitoring module monitors the operating status parameters of each submodule, including the load module, power module, fan module, liquid medicine delivery module, and spray module, in real time. The system adopts intelligent fault diagnosis technology, employing threshold judgment and anomaly detection algorithms to achieve fault detection and diagnosis and anomaly alarm processing.

[0036] (3) Deep Learning Intelligent Control: A parameter optimization system based on the target detection algorithm of the YOLOv5 architecture and the PPO reinforcement learning algorithm is used to achieve the optimal control strategy. The system deploys a lightweight algorithm model in the edge computing unit to achieve real-time intelligent decision-making, automatically adjust the nozzle angle, fan speed, and liquid flow rate according to the characteristics of the fruit trees, and realize variable spraying based on real-time environmental perception.

Claims

1. A modular orchard air-assisted spray system and quick-change device, characterized in that: include: Cloud platform, modular spray system and equipment monitoring and control system; The cloud platform provides data storage, remote access and decision support services; The modular spray system comprises: The load-bearing module provides structural support for the system and is used to fix and connect other modules; the power module provides power output for the system; the fan module includes the fan and air duct; the liquid medicine delivery module includes the liquid medicine storage tank and delivery pipeline; the spray module includes the nozzle assembly; the quick-change device includes the connector and quick-locking mechanism; The equipment monitoring and control system includes a sensor, an edge computing unit, and a local control execution unit; The sensor is connected to the edge computing unit via a communication interface, and the edge computing unit is connected to the cloud platform via wireless communication.

2. The modular orchard air-assisted spray system and quick-change device according to claim 1 is characterized in that: The cloud platform uses a deep learning algorithm to analyze orchard environmental data and equipment operation data, and the deep learning algorithm includes a target detection algorithm and a parameter optimization algorithm.

3. The modular orchard air-assisted spray system and quick-change device according to claim 1, characterized in that: The load-bearing module includes a main frame structure and a module installation interface. The main frame structure is made of lightweight materials. The module installation interface is provided with a standardized connection interface for connecting and separating with the power module, fan module, liquid medicine delivery module, and spray module through a quick-change device.

4. The modular orchard air-assisted spray system and quick-change device according to claim 1, characterized in that: The quick locking mechanism of the quick-change device adopts a snap-on connection method to achieve tool-free disassembly and assembly between modules.

5. The modular orchard air-assisted spray system and quick-change device according to claim 1, characterized in that: The sensors include a GPS locator, a fuel level sensor, a fan speed sensor, a temperature sensor, a flow sensor, a liquid medicine level sensor, an environmental monitoring sensor and an image recognition camera.

6. The modular orchard air-assisted spray system and quick-change device according to claim 1, characterized in that: The edge computing unit receives the operating parameters of each module collected by the sensor in real time, preprocesses, formats and caches the operating parameters, uploads the processed operating parameters to the cloud platform, and simultaneously receives control instructions issued by the cloud platform and forwards them to the local control execution unit.

7. The modular orchard air-assisted spray system and quick-change device according to claim 1, characterized in that: The local control execution unit includes a motor driver, a solenoid valve controller and a servo controller, which are used to execute cloud platform control instructions forwarded by the edge computing unit; when communication is interrupted, the edge computing unit performs local emergency control based on a preset security policy.

8. The modular orchard air-assisted spray system and quick-change device according to claim 2, characterized in that: The target detection algorithm is based on the YOLOv5 model and is used to identify the location, size and density of fruit trees; The total loss function of the YOLOv5 model comprehensively considers bounding box positioning, target confidence and category classification. Defined as: in, 、 、 are weight coefficients, respectively, used to balance the losses of each part; Represents the positioning loss, and the specific formula is: in, represents the intersection-over-union ratio of the predicted box and the true box, , ) represents the Euclidean distance between the center point of the predicted box and the real box, Represents the diagonal distance between the minimum enclosing rectangle of the predicted box and the real box, represents the weight coefficient, represents the aspect ratio consistency factor; Represents the confidence loss, the specific formula is: in, and represent the true and predicted confidence respectively; Represents the classification loss, and the specific formula is: in, and are the true and predicted class probabilities; The parameter optimization algorithm adopts the PPO reinforcement learning algorithm, and the steps are as follows: First, define the state space and motion space; The state dimension of the state space is expressed as: , in, Indicates fruit tree characteristic parameters, environmental parameters or equipment status parameters; The action dimension of the motion space is expressed as: , in, For fan speed adjustment; To regulate the flow of liquid medicine; Adjust the nozzle angle; Next, design the reward function. The reward function is designed as follows: in, 、 、 、 is the weight coefficient, Reward for work efficiency, Reward for liquid medicine utilization rate, Reward for equipment safety, rewards for energy consumption optimization; Then, design the loss function of the PPO reinforcement learning algorithm , the formula is: in: represents the expectation for all time steps t; A new strategy for the PPO reinforcement learning algorithm The probability ratio of taking an action at time step t with the old policy; is the advantage function estimate, which represents the difference between the action value function and the state value function in the PPO reinforcement learning algorithm; is the cropping parameter; For the trimming operation, Restricted to [ ] interval; Finally, the fruit tree information identified by the target detection algorithm is directly used as the fruit tree feature parameters of the parameter optimization algorithm, and the optimal nozzle angle, fan speed and liquid flow rate parameters in the current scene are finally output.

9. The modular orchard air-assisted spray system and quick-change device according to claim 1, characterized in that: The system adopts a three-level collaborative control architecture of cloud, edge and end: the cloud platform performs global decision optimization and issues control instructions based on deep learning algorithms, the edge computing unit is responsible for data processing, instruction forwarding and emergency local control during communication interruptions, and the local control execution unit is responsible for accurately executing control instructions and status feedback.

Images