Unmanned aerial vehicle environment monitoring and data processing method supported by edge computing

Abstract

The invention is suitable for the technical field of unmanned aerial vehicle communication, and particularly relates to an unmanned aerial vehicle environment monitoring and data processing method supported by edge computing, and the method comprises the steps: obtaining a time sequence data stream collected by a multi-source environment sensing device, and carrying out the compression and processing of the time sequence data stream based on a linear projection compression model, and obtaining a compressed feature vector; calculating a communication cost based on the current heterogeneous network condition and the energy constraint through a preset communication cost function; and executing different data processing strategies based on different communication costs. According to the method, efficient edge compression is realized in an embedded resource limited scene; resource-aware data uploading control is realized through a communication cost function; a Bayesian state fusion method based on a compressed feature space reconstructs a Kalman filtering matrix structure, effectively supports local state estimation and uncertainty control, instantly triggers early warning when an environment index crosses a boundary, and effectively enhances the risk perception and response capability of a system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicle communication, and particularly relates to an edge computing supported unmanned aerial vehicle environment monitoring and data processing method. BACKGROUND

[0002] At present, an environment monitoring system generally adopts a centralized processing architecture with cloud computing as the core. The architecture takes an unmanned aerial vehicle as an air mobile sensing platform, acquires multi-modal sensing data including temperature and humidity, gas concentration, particulate matter (PM2.5), thermal radiation and visible light image through carrying various environment sensors. The collected raw data is transmitted in real time to a remote cloud computing center through a wireless communication mode (usually 4G or 5G link), and a high-performance server completes data fusion, abnormal detection and event identification processing tasks.

[0003] In such an architecture, the flight task performed by the unmanned aerial vehicle usually depends on a static path planning made in advance, that is, the flight path points and cruise paths are predefined before the task starts. The path is generally set based on the regional map or historical task experience, and lacks the dynamic adjustment capability for sudden pollution events or temporary target changes. In addition, in order to reduce the communication and processing load, some systems introduce simple data compression, image filtering or resolution downsampling preprocessing operations on the unmanned aerial vehicle side, but still rely on the cloud to complete the core sensing judgment and decision-making process as a whole.

[0004] Such a centralized architecture has certain advantages in computing power, unified management and centralized analysis, but when facing dynamic environments, large-scale sensing areas and task scenarios with high timeliness requirements, the system performance and practicality face obvious challenges. SUMMARY

[0005] The purpose of the present application is to provide an edge computing supported unmanned aerial vehicle environment monitoring and data processing method, which aims to solve the problem that the system performance and practicality face obvious challenges when the prior art faces dynamic environments, large-scale sensing areas and task scenarios with high timeliness requirements.

[0006] The present application is implemented as follows: an edge computing supported unmanned aerial vehicle environment monitoring and data processing method, the method comprising: acquiring a time series data stream collected by a multi-source environment sensing device, and compressing and processing the time series data stream based on a linear projection compression model to obtain a compressed feature vector; calculating a communication cost based on current heterogeneous network conditions and energy constraints through a preset communication cost function; when the communication cost is higher than a preset value, a cloud reporting process is performed, the compressed feature vector is transmitted to a cloud platform, a monitoring task instruction is generated through the cloud platform, and the linear projection compression model is trained and parameter updated; When the communication cost is lower than the preset value, the observation vector is generated through the linear projection compression model, the observation matrix is ​​constructed, and it is replaced with a compressed form. The dynamic event decision engine is used to perform anomaly detection and response strategy reasoning.

[0007] Preferably, in the step of obtaining the time series data stream collected by the multi-source environmental perception device, the source of the time series data stream includes at least an onboard temperature sensor, an air quality detector, and a humidity sensor, and is represented in vector form:

[0008] in, It's a moment The state vector represents the various environmental information collected by the drone through different sensors at this moment. for The temperature detected by the onboard temperature sensor at the moment, for The air quality value detected by the air quality detector at all times, for The humidity detected by the humidity sensor at all times, represents a real vector with three rows and one column.

[0009] Preferably, in the step of compressing and processing it based on a linear projection compression model to obtain a compressed feature vector, the linear projection compression model adopts a TinyML architecture, and its transformation relationship is expressed as: ; in, It is a projection matrix with a size of 2 rows and 3 columns, which takes the input 3D features Compressed into 2D features ,right Perform linear feature transformation, , represents the bias vector, the size is 2 rows and 1 column, generating the compressed feature vector For subsequent edge collaboration tasks, Represents the compressed feature vector, which is 2 rows and 1 column in size.

[0010] Preferably, the communication cost function is expressed as:

[0011] in, It represents the communication cost, which measures the cost of the UAV to perform the communication task in the current state. The larger the value, the higher the communication cost. represents the channel congestion coefficient, Indicates the current remaining energy level of the drone, is a weight coefficient, used to control the degree of influence of channel congestion on the communication cost, the weight coefficient is used to control the degree of influence of residual energy on the communication cost.

[0012] Preferably, when the communication cost is higher than a preset value , a cloud reporting process is performed, and the compressed feature vector is transmitted to the cloud platform, the cloud platform generates a monitoring task instruction, and the linear projection compression model is trained and the parameter is updated. In the step, the cloud platform collects data from different unmanned aerial vehicle nodes, performs global fusion analysis of cross-regional environmental features, constructs a spatial distribution model and a time series evolution trend, and is used for city pollution monitoring and forest fire situation identification. The cloud platform generates a dynamic response strategy, updated model parameters or a new monitoring task instruction based on the fusion analysis result and the task scheduling demand, and sends it to the corresponding unmanned aerial vehicle or edge node through the downlink.

[0013] Preferably, when the communication cost is lower than the preset value, the linear projection compression model generates an observation vector, constructs an observation matrix, and replaces it with a compressed form in the step. Through the linear projection compression model, the multi-modal observation data is linearly compressed on the edge device to generate an observation vector with lower dimension , the observation matrix is replaced by a compressed form , wherein is the original observation matrix, representing a multi-modal, high-dimensional feature set collected by the unmanned aerial vehicle on the edge side, is a projection matrix generated by the linear projection compression model, which embeds feature dimension reduction and structure preservation information, and is used to compress high-dimensional features into low-dimensional features , is the compressed observation matrix, which is a linear dimension reduction result of , the feature dimension is smaller but still maintains the key features, and is used for subsequent reasoning or cooperative tasks.

[0014] Preferably, in the step of mapping the compressed observation vector to the real state space in the reasoning process, the state estimation formula is expressed as:

[0015] wherein: is the compressed observation feature output by the edge node, obtained by linear projection transformation of multi-modal perception data, is the observation matrix in the compressed space, is the measurement noise covariance matrix in the compressed observation space, and is the state estimation and covariance before prediction. and is the state estimation and covariance after fusion, is the Kalman gain, which determines the adjustment range of the estimation, t is the time index, the current discrete time step, represents the unmanned aerial vehicle at the time point, is the transpose matrix of .

[0016] Preferably, in the step of abnormality detection and response strategy inference by the dynamic event decision engine, the abnormality detection and response strategy inference of the target event are performed by the multi-source regularization fusion model, and the modeling form is represented as:

[0017] wherein, represents the state estimation output from the th sub-module, is the weight parameter corresponding to each sub-module, is an L2 regularization coefficient, used to suppress the weight overfitting phenomenon, is the inferred occurrence probability result of the abnormal event in the environment monitoring, is a vector composed of all weights, , is the square of the L2 norm of the overall weight vector.

[0018] Preferably, when the state estimation value exceeds the preset threshold value during the execution of the task by the unmanned aerial vehicle, the system automatically responds to the preset strategy, and the strategy includes path re-planning, cooperative node wake-up or cloud task reporting.

[0019] Preferably, the matrix is subjected to structure compression optimization through QR decomposition or singular value truncation, so that the compression operation has low computational overhead while maintaining information discriminability.

[0020] The unmanned aerial vehicle environment monitoring and data processing method supported by the edge computing provided in the application has the following beneficial effects: In an embedded resource-constrained scene, a lightweight feature dimension reduction mechanism of a TinyML model is adopted to realize efficient edge compression; A communication cost function combining channel congestion and power state is constructed to dynamically optimize the backhaul strategy and realize resource-aware data upload control; A Bayesian state fusion method based on a compressed feature space is used to reconstruct the Kalman filter matrix structure, effectively supporting local state estimation and uncertainty control; A local discrimination and response mechanism for abnormal events is constructed to trigger an early warning immediately when the environment index is out of range, effectively enhancing the risk perception and response capability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 An application environment architecture diagram of an edge computing supported unmanned aerial vehicle environment monitoring and data processing method provided by an embodiment of the present application is shown in FIG. 1. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0023] The present application mainly relates to the cross technical field of edge computing, Internet of Things and environment monitoring, specifically including unmanned aerial vehicle intelligent collaborative perception, edge side data processing and compression method, and multi-source information fusion and real-time response mechanism in complex environment. These technologies not only represent the research hotspot of current environment perception system, but also constitute the key supporting technology in the actual scene of smart city, ecological monitoring and disaster response.

[0024] Edge computing (Edge Computing) is an architecture that sinks computing resources and data processing capabilities to data sources (such as sensors or unmanned aerial vehicle platforms), aiming to reduce network transmission delay, reduce bandwidth consumption and improve system real-time processing capability. The present application significantly improves the response efficiency and task autonomy of the system to dynamic environmental changes by deploying computing tasks on the unmanned aerial vehicle body or edge node, providing a low-latency, high-efficiency solution for wide-area environment monitoring scenarios.

[0025] Internet of Things (Internet of Things, IoT) technology is a basic infrastructure system that realizes device interconnection and data collaborative sharing, and forms a dynamic perception and efficient communication task coordination system by building a heterogeneous network architecture between unmanned aerial vehicles, edge nodes and ground control platforms. In the present application, the IoT architecture supports unmanned aerial vehicle task scheduling and multi-source data fusion in a distributed multi-node environment, providing technical support for high spatio-temporal coverage and device linkage in environmental monitoring.

[0026] Environmental monitoring technology (Environmental Monitoring Technology) aims to observe and analyze air, water, land and weather elements in natural or man-made environments for a long time, continuously and with high precision. The present application takes unmanned aerial vehicles as the core platform, realizes fine collection of parameters such as pollutants, temperature and humidity, thermal radiation by integrating multi-modal sensors, and improves the recognition accuracy and monitoring robustness of key environmental change events by combining edge side data preprocessing and intelligent judgment mechanism.

[0027] Intelligent Cooperative Perception for UAVs (ICP) is one of the core innovative modules proposed in this paper. This technology leverages the DroEdge-EMS architecture to build an aerial distributed perception network using wireless short-range communication protocols (such as MAVLink) and a dynamic task allocation mechanism among multiple UAVs. The system supports real-time sharing of compressed sensing features and location information collected by each node, and combines it with lightweight inference models (such as the TinyML module) on edge nodes to perform joint analysis and event judgment on local areas. For example, when PM2.5 values ​​in a particular area are continuously abnormal, adjacent UAVs can trigger a collaborative "hotspot detection" mechanism, automatically adjusting their flight paths to increase coverage density, confirming local risks, and performing additional sampling tasks, thereby enabling rapid event response without relying on a central control node. This mechanism significantly improves environmental change perception and system robustness in complex terrain or signal-blind areas.

[0028] Edge-side data processing mechanisms are crucial for improving real-time performance and resource efficiency. To address the challenges of large amounts of data collected by drones and limited communication resources, this paper proposes an edge processing module that integrates compression encoding, feature extraction, and task-specific filtering. This module performs preliminary data filtering and encoding on the node side, effectively reducing communication overhead while preserving core information and improving the accuracy and efficiency of subsequent detection algorithms.

[0029] like Figure 1 FIG. 1 is an application environment architecture diagram of a method for drone environment monitoring and data processing supported by edge computing provided by an embodiment of the present invention, wherein the method includes: The time series data stream collected by multi-source environmental perception devices is obtained, and it is compressed and processed based on the linear projection compression model to obtain the compressed feature vector.

[0030] In this step, the perception portal of the present invention is composed of a lightweight aggregator module deployed on the drone or adjacent edge node. Its core task is to significantly reduce the communication cost and energy consumption of the original environmental data during wireless transmission without sacrificing the ability to express key features. The lightweight aggregator module receives data from the onboard temperature sensor (Temperature, )、Air Quality Detector(Air Quality, )、Humidity sensor(Humidity, ) multi-source environmental perception devices, and represents them as column vectors:

[0031] is the state vector at time , representing the multi-source environmental information collected by different sensors at that time, is the temperature detected by the on-board temperature sensor at time is the air quality value detected by the air quality detector at time is the humidity detected by the humidity sensor at time represents a three-row one-column real vector.

[0032] To achieve efficient data compression and processing on edge devices, the system introduces a linear projection compression model based on TinyML architecture, whose transformation relationship is defined as:

[0033] is the projection matrix, which is 2 rows and 3 columns, compressing the input 3-dimensional feature into 2-dimensional feature , performing linear feature transformation on , and is the bias vector, which is 2 rows and 1 column, generating the compressed feature vector for subsequent edge collaborative tasks, represents the compressed feature vector, which is 2 rows and 1 column.

[0034] This transformation projects the three-dimensional original perception vector into a low-dimensional latent space, generating the compressed feature vector for subsequent edge collaborative tasks. To ensure the balance between model expression ability and compactness, the matrix can be structurally compressed and optimized through QR decomposition or singular value decomposition (SVD), etc., so that the compression operation has low computational overhead while maintaining information recognition, suitable for low-power micro control unit (MCU) operating environment. This formula projects the original 3 features (temperature, air quality, humidity) through linear projection to get 2-dimensional features, which are used for subsequent edge collaborative computing (such as task allocation or target recognition).

[0035] The communication cost is calculated based on the current heterogeneous network conditions and energy constraints through a pre-set communication cost function. ​​​

[0036] In this step, the communication and state fusion phase is entered, i.e., the UAV-edge collaboration subsystem, which is the core intermediate layer of DroEdge-EMS, is responsible for intelligent scheduling of data backhaul decision and collaborative prediction of multi-source state information. For heterogeneous network conditions and energy constraints, the system introduces a communication cost function:

[0037] wherein, represents the congestion coefficient of the current channel (estimated by packet loss rate, delay, etc.), represents the remaining energy level of the current UAV, , is the weighted coefficient for regulating network load and energy consumption balance. This function is embedded into the backhaul strategy scheduler as an optimization objective for intelligent decision-making on whether to perform data reporting, thereby ensuring information freshness while extending flight endurance and system life cycle.

[0038] When the communication cost is higher than the preset value, the cloud reporting process of the compressed feature vector is performed to the cloud platform, the monitoring task instruction is generated through the cloud platform, and the linear projection compression model is trained and parameter updated.

[0039] In this step, to realize intelligent control of the communication scheduling strategy, the system sets an adjustable cost threshold parameter When the calculated communication cost , the system triggers the cloud reporting process of the compressed feature. The compressed feature representation is uploaded to the cloud platform or ground control center under the condition of meeting the transmission condition. This platform not only undertakes the centralized management task of feature data, but also serves as a remote intelligent hub of the DroEdge-EMS architecture, which is used to perform multiple key functions.

[0040] Firstly, the platform collects data from different UAV nodes to perform global fusion analysis of cross-regional environmental features, construct spatial distribution models and time series evolution trends, and is used for macroscopic perception in scenarios such as urban pollution monitoring and forest fire situation identification.

[0041] Secondly, the platform supports periodic evaluation and retraining of the edge model. With the support of centralized computing resources, the model is optimized in accuracy, adjusted in parameters, or learned in abnormal features to improve subsequent inference ability.

[0042] ​Thirdly, based on the fusion analysis results and task scheduling requirements, the cloud can generate dynamic response strategies, updated model parameters or new monitoring task instructions, and send them to the corresponding UAV or edge node through the downlink. This feedback process constitutes a closed-loop mechanism in the DroEdge-EMS system from environmental perception, state uploading, central optimization to local feedback execution, realizing the intelligent regulation and control cycle of "perception-decision-feedback-re-perception", and significantly enhancing the system's task adaptive ability and evolutionary optimization ability in long-term operation.

[0043] When the communication cost is lower than the preset value, the observation vector is generated by the linear projection compression model, the observation matrix is constructed, and it is replaced by the compressed form, and the dynamic event decision engine is used for abnormal detection and response strategy reasoning.

[0044] In this step, if the current communication cost is higher than the threshold value, the data is retained locally, and the edge side state estimation and reasoning path is entered. The threshold can be dynamically adjusted according to the task urgency, historical network performance or flight cycle, so as to balance the scheduling between energy efficiency, delay and monitoring quality.

[0045] At the state estimation level, the DroEdge-EMS system introduces an improved Kalman filter mechanism (KF) suitable for edge compression feature space. Unlike the traditional KF, which estimates on the original high-dimensional sensor data, the system first uses the TinyML module to perform linear compression on the multi-modal observation data on the edge device, generating a lower-dimensional observation vector , and then performs state fusion in this compressed space. This processing paradigm fully embodies the advantage of edge computing "intelligence on site", that is, completing feature compression, state prediction and anomaly identification locally, greatly reducing communication pressure and computational overhead.

[0046] To adapt to the state reasoning requirements of the compressed space, the system replaces the original observation matrix with the compressed form , where is the projection matrix generated by the TinyML compression model. This matrix embeds the information of feature dimension reduction and structure preservation, so that the compressed observation vector can still be effectively mapped to the real state space in the reasoning process, is the original observation matrix, representing the multi-modal, high-dimensional feature set collected by the UAV at the edge side, is the projection matrix generated by the linear projection compression model, used to compress the high-dimensional feature into a low-dimensional feature , is the compressed observation matrix, which is a replacement for The linear dimension reduction result of the feature dimension is smaller but still retains the key features for subsequent reasoning or collaborative tasks.

[0047] The state estimation formula is as follows: ; Wherein: is the compressed observation feature output by the edge node, which is obtained by linear projection transformation of multi-modal perception data, is the observation matrix in the compressed space, is the measurement noise covariance matrix in the compressed observation space, and is the state estimation and covariance before prediction, and is the fused state estimation and covariance, is the Kalman gain, which determines the estimation adjustment amplitude, t is the time index, the current discrete time step, indicates the unmanned aerial vehicle at the time point, is the transpose matrix of .

[0048] The mechanism runs on the edge device with an estimation period of 200ms, supporting real-time state tracking in dynamic environments. With the help of information fusion capability in the compressed space, DroEdge-EMS not only improves the processing performance of the edge node, but also guarantees the accuracy of state estimation under compression constraints, providing efficient intelligent reasoning capability for autonomous monitoring tasks in complex environments.

[0049] The upper layer of the system is a dynamic event decision engine, which constitutes the intelligent control center of the "perception-fusion-response" closed loop in DroEdge-EMS. This module aggregates the state estimation results and their uncertainty representation uploaded by the edge layer, and performs abnormal event detection and response strategy reasoning through a multi-source regularization fusion model. Its modeling form is as follows:

[0050] Wherein, indicates the state estimation output from the th sub-module, is the weight parameter corresponding to each sub-module, is the L2 regularization coefficient, used to suppress the weight overfitting phenomenon, is the inferred occurrence probability result of the abnormal event in environmental monitoring, is the vector composed of all weights, , is the square of the L2 norm of the overall weight vector.

[0051] In terms of emergency response, the module embeds an emergency triggering mechanism based on logical rules. For example, in the air quality anomaly detection task, when the state estimation value exceeds the AQI threshold set by the World Health Organization (WHO) , the system will trigger the following actions:

[0052] This decision logic can be extended to other parameters (such as sudden temperature rise, sudden humidity drop, etc.) and support response strategy configuration, including path re-planning, coordinated node wake-up, and cloud (i.e., cloud platform) task reporting, thereby realizing self-organization, self-repair, and scenario adaptation of unmanned systems.

[0053] In summary, the present application constructs a closed-loop unmanned monitoring system that runs through multi-modal environmental perception, edge computing processing, and intelligent decision reasoning. Its core advantages are reflected in the following aspects: Firstly, the system can complete feature compression and preliminary perception judgment on the edge side (such as the unmanned aerial vehicle body or near-ground edge node), effectively reducing the dependence on the cloud and significantly improving the real-time performance and task autonomy level of the system. Secondly, by integrating a lightweight state estimation mechanism (such as a Kalman filter), DroEdge-EMS still has excellent robustness and prediction accuracy in the face of high dynamic and strong uncertainty environments, ensuring the time series continuity and estimation reliability of key state variables.

[0054] In addition, the communication cost modeling strategy embedded in the system can consider the wireless channel state and energy remaining information to achieve dynamic optimization of communication strategies under resource constraints, thereby prolonging the system task cycle and alleviating the transmission bottleneck. Finally, the introduction of the event-driven mechanism enables the system to have the ability to respond quickly to sudden risk factors. Through dynamic evaluation and threshold triggering judgment of monitoring variables, the system can automatically complete early warning push, task rescheduling, or multi-unmanned aerial vehicle coordinated response when detecting potential environmental anomalies (such as air quality exceeding standards, sudden temperature changes, etc.), demonstrating high intelligence and situational adaptability.

[0055] Overall, the present application takes edge intelligence as the core driver, integrates key mechanisms such as perception compression, state prediction, resource optimization, and event reasoning, and exhibits strong adaptability, scalability, and operational efficiency in complex environmental monitoring tasks. This architecture represents a "end-edge-coordination-reasoning" paradigm in intelligent unmanned systems, providing theoretical support and technical foundation for promoting the intelligence, low power consumption, and high reliability of the next generation of autonomous environmental monitoring systems.

[0056] Compared with the traditional architecture relying on cloud centralized computing, the present application is based on the edge computing paradigm, and the preprocessing and feature compression of multi-modal environment data are completed in the unmanned aerial vehicle body or near-ground edge node, which effectively reduces the data back transmission frequency and bandwidth occupation, and the transmission demand is reduced by about 60% compared with the original scheme. The system realizes local intelligent inference through the embedded TinyML module, significantly reduces the model parameter size and operation energy consumption, and improves the deployment adaptability in the resource limited scene. The traditional architecture usually hands over the event response to the remote server for processing, resulting in high decision delay, while the DroEdge-EMS system integrates state estimation and event identification mechanism, supports real-time response and task scheduling on the unmanned aerial vehicle side, and improves the real-time performance and task continuity of the system. In terms of processing complexity, compared with the large parameter and computationally intensive Transformer model, DroEdge-EMS realizes efficient operation through lightweight structure, is suitable for large-scale deployment and long-term running of intelligent environment monitoring tasks, and has good scalability and task robustness.

[0057] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. The edge computing-supported UAV environmental monitoring and data processing method is characterized by: The method comprises: Obtain the time series data stream collected by multi-source environmental perception devices, compress and process it based on the linear projection compression model to obtain the compressed feature vector; The communication cost is calculated based on the current heterogeneous network conditions and energy constraints through a preset communication cost function; When the communication cost is higher than the preset value, the cloud reporting process is executed to transmit the compressed feature vector to the cloud platform, generate monitoring task instructions through the cloud platform, and train and update the parameters of the linear projection compression model; When the communication cost is lower than the preset value, the observation vector is generated through the linear projection compression model, the observation matrix is ​​constructed, and it is replaced with a compressed form. The dynamic event decision engine is used to perform anomaly detection and response strategy reasoning.

2. The edge computing-supported drone environment monitoring and data processing method according to claim 1, characterized in that: In the step of obtaining the time series data stream collected by the multi-source environmental perception device, the source of the time series data stream includes at least an onboard temperature sensor, an air quality detector, and a humidity sensor, and is represented as a vector: ； in, It's a moment The state vector represents the various environmental information collected by the drone through different sensors at this moment. for The temperature detected by the onboard temperature sensor at the moment, for The air quality value detected by the air quality detector at all times, for The humidity detected by the humidity sensor at all times, represents a real vector with three rows and one column.

3. The edge computing-supported drone environment monitoring and data processing method according to claim 2, characterized in that: In the step of compressing and processing it based on the linear projection compression model to obtain the compressed feature vector, the linear projection compression model adopts the TinyML architecture, and its transformation relationship is expressed as: ； It is a projection matrix with a size of 2 rows and 3 columns, which takes the input 3D features Compressed into 2D features ,right Perform linear feature transformation, , represents the bias vector, the size is 2 rows and 1 column, generating the compressed feature vector For subsequent edge collaboration tasks, Represents the compressed feature vector, which is 2 rows and 1 column in size.

4. The edge computing-supported drone environment monitoring and data processing method according to claim 1, characterized in that: The communication cost function is expressed as: ； in, It represents the communication cost, which measures the cost of the UAV to perform the communication task in the current state. The larger the value, the higher the communication cost. represents the channel congestion coefficient, Indicates the current remaining energy level of the drone, is the weight coefficient, which is used to control the impact of channel congestion on communication cost. Used to control the impact of residual energy on communication cost.

5. The edge computing-supported drone environment monitoring and data processing method according to claim 1, characterized in that: When the communication cost Higher than the preset value When the compressed feature vector is transmitted to the cloud platform, the cloud platform generates monitoring task instructions, and trains and updates the parameters of the linear projection compression model. The cloud platform aggregates data from different drone nodes to perform a global fusion analysis of cross-regional environmental characteristics, builds a spatial distribution model and time series evolution trend, which is used for urban pollution monitoring and forest fire situation identification. Based on the fusion analysis results and task scheduling requirements, the cloud platform generates dynamic response strategies, updated model parameters or new monitoring task instructions, and sends them to the corresponding drones or edge nodes through the downlink.

6. The edge computing-supported drone environment monitoring and data processing method according to claim 1, characterized in that: When the communication cost is lower than the preset value, the observation vector is generated by the linear projection compression model, the observation matrix is ​​constructed, and it is replaced with the compressed form. The multimodal observation data is linearly compressed on the edge device through the linear projection compression model to generate a lower-dimensional observation vector. , the observation matrix Replace with compressed form ,in, is the original observation matrix, which represents the multimodal, high-dimensional feature set collected by the UAV at the edge. is the projection matrix generated by the linear projection compression model, which is used to convert high-dimensional features Compress into low-dimensional features , is the compressed observation matrix.

7. The edge computing-supported drone environment monitoring and data processing method according to claim 6, characterized in that: In the step of mapping the compressed observation vector to the real state space during the inference process, the state estimation formula is expressed as: ； in: It is the compressed observation feature output by the edge node, which is obtained by linear projection transformation of multimodal perception data. is the measurement matrix in the compressed space, is the measurement noise covariance matrix in the compressed observation space, and is the state estimate and covariance before prediction, and is the fused state estimate and covariance, is the Kalman gain, which determines the estimated adjustment amplitude, t is the time index, and the current discrete time step indicates the time point at which the drone is present. yes The transposed matrix of .

8. The edge computing-supported UAV environmental monitoring and data processing method according to claim 6, characterized in that: In the step of anomaly detection and response strategy reasoning through the dynamic event decision engine, anomaly detection and response strategy reasoning of target events are performed through a multi-source regularization fusion model. The modeling form is expressed as: ； in, Indicates that it comes from The state estimation output of each submodule is: is the weight parameter corresponding to each submodule, is the L2 regularization coefficient, which is used to suppress the over-fitting phenomenon of weights. It is the inferred probability result of abnormal events in environmental monitoring. is a vector of all weights, , is the squared L2 norm of the overall weight vector.

9. The edge computing-supported UAV environment monitoring and data processing method according to claim 8, characterized in that: During the UAV's mission, when the state estimation value When the preset threshold is exceeded, the system automatically responds to the preset strategy, which includes path replanning, collaborative node wake-up or cloud task reporting.

10. The edge computing-supported drone environment monitoring and data processing method according to claim 3, characterized in that: matrix Structural compression is optimized through QR decomposition or singular value truncation, so that the compression operation has low computational overhead while maintaining information recognition.

Images