Edge cloud cooperative natural reserve ecological monitoring and emergency response system

Abstract

The application discloses an edge-cloud collaborative natural reserve ecological monitoring and emergency response system in the technical field of ecological protection, which comprises an edge terminal, an edge-cloud collaborative transmission module and a cloud terminal. The edge terminal is used for collecting natural reserve ecological monitoring data, performing data preprocessing and preliminary identification of abnormalities, and obtaining key information. The edge-cloud collaborative transmission module is used for transmitting the key information. The cloud terminal is used for performing global risk analysis and evaluation according to a model corresponding to the key information abnormal event, constructing an ecological situation atlas of the natural reserve, and generating an emergency disposal instruction. The emergency dispatch execution module is used for sending the emergency disposal instruction and displaying the ecological situation atlas. The lightweight inference module is pre-deployed on the edge terminal through hierarchical computing power decoupling, the ecological monitoring data is locally preprocessed and preliminarily screened, the cloud terminal is responsible for multi-scene abnormality research and global situation analysis by concentrating computing power, and the edge real-time response and the cloud intelligent decision-making efficient cooperation are improved.

Description

Technical Field

[0001] This invention relates to the field of ecological protection, specifically to the ecological monitoring and emergency response system for the Bianyun Collaborative Nature Reserve. Background Technology

[0002] For national nature reserves that are vast, have complex terrain, and limited communication conditions, existing monitoring methods mainly rely on manual inspections, isolated electronic devices, and simple IoT sensors. These methods generally suffer from problems such as large data transmission delays, waste of computing resources, delayed response to anomalies, and chaotic global scheduling.

[0003] Specifically, due to insufficient computing power of edge devices, high-throughput data such as high-definition video cannot be processed in real time. Furthermore, the centralized inference model, which transmits all raw data back to the cloud, is limited by bandwidth bottlenecks in areas with weak or no network access, resulting in significant delays in detecting emergencies such as illegal intrusions and fire hazards. In addition, traditional solutions lack an automated closed-loop mechanism from monitoring and identification to early warning response and review. From anomaly detection to task scheduling, they heavily rely on manual judgment and telephone notifications, making rapid and effective global intervention in the early stages of an incident difficult. Summary of the Invention

[0004] The purpose of this invention is to provide an edge-cloud collaborative ecological monitoring and emergency response system for nature reserves, which solves the problem that existing ecological monitoring systems are unable to achieve integrated real-time response at the edge, intelligent decision-making in the cloud, and closed-loop emergency response for the special environment of nature reserves.

[0005] The present invention achieves the above objectives through the following technical solutions: The Bianyun Collaborative Nature Reserve Ecological Monitoring and Emergency Response System includes: At the edge, ecological monitoring data of the protected area is collected, preprocessed and initially identified as anomalies are performed on the ecological monitoring data, the type of abnormal event is determined, and key information is obtained and uploaded. The key information includes the type of abnormal event, the coordinates of the event location, and the ecological monitoring data. The edge-cloud collaborative transmission module is used to receive and upload the key information. In the cloud, it is used to determine the judgment model based on the type of abnormal event according to the received key information, conduct global risk analysis and assessment, generate emergency response instructions, and construct the ecological status map of the protected area; The emergency dispatch execution module has a task dispatch and tracking module and multiple patrol terminals. It is used to receive emergency response instructions and send emergency response instructions to the patrol terminals according to the type of abnormal event and the coordinates of the incident location, and to set up alarms and display ecological situation maps.

[0006] As a further optimization of the invention, the edge-cloud collaborative transmission module includes a communication controller, which adopts an adaptive communication protocol. The adaptive communication protocol dynamically switches between wired communication, wireless communication, and satellite communication modes based on the network signal strength within the nature reserve.

[0007] As a further optimization of the invention, the edge-cloud collaborative transmission module also includes a data encryption module, which is used to encrypt and transmit the key information to the cloud.

[0008] As a further optimization of the invention, the monitoring module consists of distributed infrared cameras, drone inspection terminals, environmental sensors, and acoustic monitoring equipment.

[0009] As a further optimization of the invention, when the edge terminal determines that the type of abnormal event is illegal intrusion: key information is obtained based on the infrared camera; the cloud calls the illegal intrusion judgment model, generates a reference location, outputs a comprehensive risk score, and generates an illegal intrusion emergency response command based on the comprehensive risk score and the reference location; the emergency dispatch execution module dispatches patrol terminals and drone inspection terminals to search for illegal intruders based on the illegal intrusion emergency response command.

[0010] As a further optimization of the invention, when the edge terminal determines that the type of abnormal event is a fire hazard: key information is obtained based on the infrared camera and environmental sensors; the cloud calls the fire hazard assessment model to predict the fire spread area, outputs a comprehensive risk score, and generates a fire hazard emergency response command based on the fire spread area and the comprehensive risk score; the emergency dispatch execution module dispatches patrol terminals and drone inspection terminals to carry out fire fighting and personnel evacuation based on the fire hazard emergency response command.

[0011] As a further optimization of the invention, when the edge terminal determines that the type of abnormal event is species abnormality: key information is obtained based on infrared cameras, acoustic monitoring equipment, and drone inspection terminals; the cloud calls the species abnormality judgment model to generate species abnormality coordinates, outputs a comprehensive risk score, and generates species abnormality emergency response instructions based on the species abnormality coordinates and comprehensive analysis score; the emergency dispatch execution module dispatches patrol terminals and drone inspection terminals based on the species abnormality emergency response instructions to perform sampling tasks and verification at the event location.

[0012] As a further optimization of the invention, when the edge terminal determines that the type of abnormal event is habitat destruction: key information is obtained based on the drone inspection terminal and environmental sensors; the cloud calls the habitat destruction assessment model to generate verification coordinates and habitat change type, outputs a comprehensive risk score, and generates a habitat destruction emergency response instruction based on the verification coordinates, habitat change type and comprehensive risk score; the emergency dispatch execution module dispatches the patrol terminal and drone inspection terminal to the verification coordinates to collect evidence based on the habitat destruction emergency response instruction.

[0013] The ecological monitoring and emergency response methods for the Bianyun Collaborative Nature Reserve include the following steps: Step 1: The edge device collects multi-source ecological monitoring data, performs data preprocessing and initial anomaly identification, and filters out key information; Step 2: The edge-cloud collaborative transmission module uploads the key information identified in Step 1 to the cloud; Step 3: The cloud gathers key information from multiple edge devices, calls the multi-scenario anomaly analysis module to perform global anomaly analysis and risk level assessment, constructs an ecological status map of the protected area, and generates emergency response instructions; Step 4: The emergency dispatch execution module receives the emergency response instruction, executes patrol resource dispatch, pushes the response task, alarms, and displays the ecological situation map.

[0014] As a further optimization of the invention, step five is also included: the emergency dispatch execution module tracks the progress of task handling in real time and transmits the on-site verification data back to the cloud, which then updates the ecological situation map and generates a review file.

[0015] The beneficial effects of this invention are as follows: 1) This invention deploys a lightweight inference module at the edge by decoupling computing power in a layered manner, which performs local real-time cleaning, noise reduction and compression, and anomaly screening on ecological monitoring data; while the cloud concentrates computing power to be responsible for anomaly judgment in multiple scenarios and global situation analysis, breaking the computing power barrier between the edge and the cloud in the traditional solution. While reducing the amount of invalid data transmission and bandwidth occupation, it realizes efficient division of labor and collaboration between real-time edge response and intelligent cloud decision-making. 2) This invention constructs a fully automated closed-loop emergency response mechanism that integrates "monitoring-identification-early warning-response-review". For various emergencies such as illegal intrusion, habitat destruction, species abnormalities, and fire hazards, the system has preset multi-level judgment models and emergency instructions. It can automatically complete the entire process from event discovery, cloud-based comprehensive risk scoring, generation of disposal instructions to patrol task push within seconds, and conduct full tracking and review of the disposal. This changes the lagging mode of traditional manual judgment and telephone dispatch, and realizes second-level response and efficient disposal of emergencies. 3) This invention achieves secure fusion and reliable transmission of multi-source heterogeneous data through adaptive communication protocols, data encryption modules, and local caching mechanisms in the edge-cloud collaborative transmission module. It can dynamically switch between wired, wireless, and satellite communication modes, and encrypt and cache key data locally, ensuring the real-time performance, security, and global availability of data in complex network environments. At the same time, through the ecological situation map, managers can fully grasp the real-time situation of the protected area, providing support for intelligent and refined management decisions. Attached Figure Description

[0016] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a schematic diagram of the edge end structure of the present invention; Figure 3 This is a schematic diagram of the lightweight inference module structure of the present invention; Figure 4 This is a schematic diagram of the edge-cloud collaborative transmission module structure of the present invention; Figure 5 This is a schematic diagram of the cloud structure of the present invention; Figure 6 This is a schematic diagram of the multi-scenario anomaly assessment model library structure of the present invention; Figure 7 This is a schematic diagram of the emergency dispatch execution module structure of the present invention; Figure 8 This is a flowchart of the present invention; In the diagram: 100, Edge terminal; 200, Edge-cloud collaborative transmission module; 300, Cloud terminal; 400, Emergency dispatch execution module; 110, Infrared camera; 120, UAV inspection terminal; 130, Environmental sensor; 140, Acoustic monitoring equipment; 150, Lightweight inference module; 151, Data cleaning unit; 152, Target initial identification unit; 153, Compression encoding unit; 210, Communication controller; 220, Data encryption module; 230, Local cache module; 310, Multi-source data fusion module; 320, Multi-scenario anomaly analysis module; 321, Illegal intrusion model; 322, Fire hazard model; 323, Species anomaly model; 324, Habitat destruction model; 340, Emergency command generation module; 330, Emergency command generation module; 340, Map generation module; 410, Task scheduling and tracking module; 420, Patrol terminal; 430, Monitoring screen; 440, Early warning broadcasting equipment. Detailed Implementation

[0017] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0018] First Embodiment like Figures 1 to 7 As shown, this embodiment relates to an edge-cloud collaborative ecological monitoring and emergency response system for nature reserves. This system is particularly suitable for key ecological functional areas with large patrol areas, high patrol difficulty, and the need for 24-hour uninterrupted monitoring. These areas are mostly national nature reserves and world natural heritage sites with an area of ​​over 100 square kilometers and rugged terrain, typically covering mountains, dense forests, wetlands, etc., and most areas lack public mobile communication network coverage.

[0019] This system achieves real-time data cleaning and initial anomaly screening by deploying lightweight AI-enabled edge nodes at the front end; it solves transmission challenges in complex terrain through adaptive communication protocols; it constructs multi-scenario analysis models and a global situational map in the cloud to achieve intelligent decision-making and resource scheduling; and finally, through a closed-loop emergency execution layer, it translates decisions into actions and tracks them, thereby achieving all-weather, full-coverage, and automated intelligent patrolling of protected areas. The system includes the following modules: Edge terminal 100, deployed at the monitoring site, includes a monitoring module composed of distributed infrared cameras 110, drone inspection terminals 120, environmental sensors 130, and acoustic monitoring equipment 140, used to collect multi-source ecological monitoring data of the protected area. The edge terminal also includes a lightweight inference module 150, which performs data preprocessing, preliminary anomaly identification, and compression encoding on the multi-source ecological monitoring data to form key information, and uploads the identified key information. Data preprocessing includes data cleaning. The lightweight inference module 150 has lightweight models corresponding to illegal intrusion, fire hazards, species anomalies, and habitat destruction.

[0020] The edge-cloud collaborative transmission module 200 includes a communication controller 210, a data encryption module 220, and a local cache module 230. The edge-cloud collaborative transmission module 200 communicates with the edge detection and sensing module 100 and the cloud 300 through the communication controller 210. The data encryption module 220 encrypts key information to form encrypted data. The communication controller 210 uses an adaptive communication protocol to upload the encrypted data to the cloud 300. This adaptive communication protocol dynamically switches between wired, wireless, and satellite communication modes based on the network signal strength within the nature reserve to achieve encrypted data interaction, computing power task allocation, and control command issuance between the edge and the cloud 300. The local cache module 230 is used to cache key information uploaded by the edge device 100.

[0021] The cloud-based module 300, deployed in the monitoring center, includes a multi-data fusion module 310, a multi-scenario anomaly analysis module 320, and a map generation module 330. The multi-data fusion module 310 aggregates key information uploaded by the edge-cloud collaborative transmission module 200. The multi-scenario anomaly analysis module 320 stores analysis models for multiple anomaly scenarios. Based on the preliminary anomaly identification results, the multi-scenario anomaly analysis module 320 calls the corresponding anomaly scenario analysis model to perform global anomaly analysis and risk level assessment, generating standardized emergency response instructions. The map generation module 330 calls the analysis models to construct an ecological status map of the protected area based on the global anomaly analysis and risk level assessment results.

[0022] The emergency dispatch execution module 400 includes a task dispatch and tracking module 410, a patrol terminal 420, a monitoring screen 430, and an early warning broadcasting device 440. The task dispatch and tracking module 410 receives and executes emergency response instructions, ecological status maps of the protected area, and global anomaly analysis and risk level assessment results issued by the cloud-based system 300. It connects with the patrol terminal 420, monitoring screen 430, and early warning broadcasting device 440 to automatically dispatch patrol resources, push response tasks, and send early warning information, achieving a closed-loop response to emergencies. The patrol terminal 420 is carried by patrol personnel and is used to receive and push response tasks. The monitoring screen 430 displays the global anomaly analysis and risk level assessment results, as well as the ecological status map of the protected area, at the monitoring center. The early warning broadcasting device 440 receives early warning information and continuously plays voice warnings.

[0023] The edge-cloud collaborative nature reserve ecological monitoring and emergency response system can achieve a closed-loop operation from edge detection to cloud-based decision-making and patrol interception. The following sections will use examples of illegal intrusion, fire hazards, species anomalies, and habitat destruction to introduce the system's specific workflow.

[0024] The workflow based on unauthorized intrusion is as follows: (1) Edge detection and sensing: Edge monitoring nodes are deployed in a grid pattern at the main entrances of the protected area, along historically high-risk poaching routes, and along the borders with surrounding communities. For example, they are spaced 500-1000 meters apart, depending on the terrain. Each node consists of the following components: The infrared camera 110 is an outdoor-specific infrared trigger camera with a 4G / BeiDou communication module. The camera resolution is no less than 4 million pixels and it has night vision capabilities.

[0025] The lightweight inference module 150 is connected to the infrared camera 110. The lightweight inference module 150 performs data preprocessing, initial target identification, and compression encoding based on low-power AI chips such as the Rockchip RV1126. The lightweight inference module 150 consists of a data cleaning unit 151, an initial target identification unit 152, and a compression encoding unit 153.

[0026] The workflow of the lightweight inference module 150 is as follows: When the infrared camera 110 is triggered by the heat of an animal or human body, it captures a 10-second 1080P / 30fps video and records metadata such as ambient light and temperature.

[0027] The data cleaning unit 151 first performs differential analysis. If the video shows very little change across multiple consecutive frames (i.e., a pixel change rate of less than 5%), the video is determined to be invalid data and discarded without further processing. This step filters out a large number of invalid triggers.

[0028] The target initial identification unit 152 loads a YOLOv5s target detection model that has been cropped using 8-bit integer quantization for the cleaned video. This model runs on the RV1126 chip and achieves an inference speed of up to 25 FPS. The model detects two types of targets: "people" and "vehicles." If a "person" or "vehicle" is detected in any frame with a confidence level greater than 0.6, it is classified as a "suspected illegal intrusion" event, thus achieving preliminary anomaly identification.

[0029] When a suspected unauthorized intrusion occurs, the compression encoding unit 153 immediately uses a hardware encoder to compress the original 10-second video (approximately 50MB) into a key segment in H.265 format in real time. Simultaneously, the compression encoding unit 153 packages structured information such as the type, confidence level, location coordinates, and timestamp of the abnormal event to form key information about the unauthorized intrusion. The edge detection and perception module 100 then sends this key information to the communication controller 210.

[0030] (2) Edge-cloud collaborative transmission: The communication controller 210 of the edge-cloud collaborative transmission module 200 continuously monitors the network status. Most of the protected area lacks 4G signal. The transmission mechanism is as follows: Initial attempt: The communication controller 210 first attempts to connect to the cloud 300 via the 4G module. If the connection is successful within 5 seconds, communication with the cloud 300 is initiated via the 4G module. If a 5-second timeout occurs, it indicates a connection failure, and the network is deemed unavailable.

[0031] State Switching: According to a preset strategy, after a connection failure, the communication controller 210 automatically switches to "satellite transmission mode." It simplifies the previously generated key information about illegal intrusion into a message of approximately 100 bytes, containing "timestamp, coordinates, and type of abnormal event," and sends it to the BeiDou command center via the connected BeiDou short message module. The BeiDou command center then forwards the message to the cloud 300. For example, the text indicating the type of abnormal event can be edited to "suspected illegal intrusion." The compressed video data is decompressed and stored in the local cache module 220, which has a capacity of 128GB and can continuously store over 72 hours of key event video.

[0032] Resume Transmission After Network Disconnection: When patrol personnel carry mobile base stations or weather changes cause weak 4G signals, the communication controller 210 restores the 4G connection and automatically initiates a "resume transmission request" to the cloud 300. After the cloud 300 responds, video data in the local cache module 220 that has not yet been synchronized to the cloud 300 is sent to the cloud 300 via an encrypted transmission channel using a resume transmission method. The encrypted transmission channel is established by the data encryption module 220 and uses TLS 1.3 encryption. TLS 1.3 is a transport layer security protocol.

[0033] Through the above mechanism, even in areas with no signal, key alarm information can be delivered via satellite within seconds, ensuring that no core information is missed and that video evidence is transmitted back later.

[0034] (3) Cloud-based intelligent decision-making: Cloud 300 is deployed on a server equipped with a GPU in the monitoring center. The specific working process of Cloud 300 is as follows: ①Information aggregation: The multi-source data fusion module 310 receives encrypted data from multiple edge-cloud collaborative transmission modules 200 via a message queue, decrypts the encrypted data to obtain decrypted data, and sends the decrypted data to the multi-scenario anomaly analysis module 320. This decrypted data is the key information before encryption.

[0035] ②Multi-scenario anomaly analysis: The multi-scenario anomaly analysis module 320 incorporates a multi-scenario anomaly analysis model. Based on the preliminary anomaly identification results, the multi-scenario anomaly analysis module 320 automatically invokes the illegal intrusion model 321 for anomaly analysis. The illegal intrusion model 321 is a graph neural network-based inference engine trained on historical data. Its inputs include: The current location coordinates and trigger time of the "suspected unauthorized intrusion" event; Centered on the coordinates of the current "suspected illegal intrusion" event, and covering the adjacent edge monitoring nodes, the event triggering frequency and spatiotemporal distribution over the past 72 hours; Authorized access records of entrance and exit gates in the area adjacent to the "suspected illegal intrusion" incident during the current time period.

[0036] The illegal intrusion model 321 outputs a comprehensive risk score based on the above input data. Historical data includes the above input data from different time periods within the protected area, as well as the results of identifying "suspected illegal intrusion" events. Within the neighboring area, the higher the frequency of "suspected illegal intrusion" events triggered in the past 72 hours, the higher the comprehensive risk score. If there are authorized passage records at entrance / exit gates during the time period of the "suspected illegal intrusion," the comprehensive risk score is reduced.

[0037] ③ Emergency command generation: In this embodiment, when the comprehensive risk score reaches 0.8, it indicates the presence of an unauthorized intruder. The infrared camera 110 monitors the intruder in real time, and after identifying the intruder, determines reference coordinates based on the position of the intruder's human image in the image data. The emergency command generation module 340 is immediately triggered, generating an unauthorized intrusion emergency response command in JSON format based on the reference coordinates.

[0038] ④ Atlas update: Based on the anomaly analysis and risk assessment results from multiple scenarios, the map generation module 320 generates an ecological status map. This map, based on a GIS engine, overlays protected area zoning, real-time locations of patrol personnel, and road network layers, marking reference coordinates on the layers and displaying the status of event handling.

[0039] (4) Emergency dispatch: After receiving the emergency response command for unauthorized intrusion, the task scheduling and tracking module 410 performs the following automatic operations: Resource matching: The task scheduling and tracking module 410 queries the location information of the patrol terminal 420 to find the patrol personnel closest to the incident site. For example, if it finds that the patrol personnel is 1.2 kilometers away from the incident site, it then queries the available drone inspection terminals 120 in the drone hangar.

[0040] Task Push: The task scheduling and tracking module 410 pushes the emergency response instructions for illegal intrusion to the App on the patrol terminal 420 of the patrol personnel. The App vibrates and broadcasts a voice alert to the patrol personnel. Simultaneously, the task scheduling and tracking module 410 sends instructions to multiple drone inspection terminals 120 to patrol towards the reference coordinates. The reference coordinates are updated in real time based on the monitoring of the infrared camera 110. After the multiple drone inspection terminals 120 reach the reference coordinates, if the reference coordinates have not been updated, the multiple drone inspection terminals 120 will patrol around the reference coordinates. Each drone inspection terminal 120 is equipped with a camera and searches for illegal intruders based on human recognition. Once a drone inspection terminal 120 finds an illegal intruder, it follows the intruder's movement, using its own position coordinates as the intruder's coordinates, and sends the intruder's coordinates to the task scheduling and tracking module 410.

[0041] Information dissemination: The monitoring screen 430 issues an alarm and displays an ecological status map. At the same time, the system uses the nearest warning broadcasting device 440 to repeatedly play a voice warning: "You have entered the core area of ​​the nature reserve. Please return immediately via the original route, otherwise you will be dealt with according to law."

[0042] On-site handling and closed-loop tracking: Before the drone inspection terminal 120 locates the illegal intruder, the task scheduling and tracking module 410 sends reference coordinates to the patrol terminal 420. After the drone inspection terminal 120 locates the illegal intruder, the task scheduling and tracking module 410 sends the intruder's coordinates to the patrol terminal 420. The patrol terminal 420 is equipped with an app for issuing patrol tasks. The app displays the reference coordinates and the intruder's coordinates. After the patrol personnel click "Acknowledge" on the app, the system starts timing and tracking. After arriving at the scene, the patrol personnel take photos and videos of the scene using the app and upload the handling situation. In this embodiment, the patrol personnel report the handling situation via SMS. For example, the patrol personnel can report the handling situation by uploading an SMS with the content "An illegal hiker has been intercepted and is being taken away." The task scheduling and tracking module 410 receives these feedbacks in real time and updates them on the monitoring screen. Based on the feedback, the cloud changes the handling status of the event in the situation map from "In Handling" to "Pending Review".

[0043] After the incident is completely resolved, the system automatically packages all data into an immutable "handling review file." This data includes trigger videos, satellite alarms, analysis logs, dispatch records, handling photos, and personnel trajectories. This allows monitoring center personnel to review the file and analyze the cause of the intrusion incident, such as whether there was any fence breach. Based on this analysis, the deployment locations of edge nodes can be optimized or emergency plans modified, truly achieving adaptive and evolving closed-loop management.

[0044] The workflow based on fire hazard is as follows: (1) Edge detection and sensing: Infrared cameras 110 and environmental sensors 130 are deployed in the fire protection zone. The environmental sensors 130 include smoke sensors, temperature sensors, humidity sensors, and wind speed and direction sensors. Both infrared cameras 110 and environmental sensors 130 are used for fire detection. Infrared cameras 110 are also infrared-triggered cameras, which are triggered when the ambient temperature exceeds a temperature threshold. At the same time, environmental sensors 130 detect the ambient temperature and smoke concentration in real time.

[0045] The workflow of the lightweight inference module 150 is as follows: Environmental sensor 130 continuously collects temperature and smoke concentration data. Data cleaning unit 151 smooths the raw signal using a sliding window, filtering out transient spikes caused by fog and dust. If the smoke concentration remains above a smoke concentration threshold for a first set time, and the relative temperature increase in multiple consecutive temperature measurements exceeds a set temperature, the system enters the preliminary identification stage. The first set time is preferably 2 seconds. The relative temperature increase is the difference between the temperature measured by one temperature sensor and the average temperature measured by multiple other temperature sensors. In this embodiment, the number of temperature measurements is 3, and the set temperature is preferably 5°C.

[0046] During the initial identification process, the target initial identification unit 152 loads a lightweight fire hazard model. This model is preferably the YOLOv5s model, which detects two types of targets: "flame" and "smoke". If the confidence score of flame is greater than 0.7 or the confidence score of smoke is greater than 0.75 in any frame, it is determined to be a "suspected fire" event.

[0047] After a suspected fire occurs, the compression encoding unit 153 immediately extracts the key frame of the thermal imaging and the first 5 seconds of the visible light video, compresses them into H.265 format, and packages the structured data such as the type of abnormal event, confidence level, flame / smoke area ratio, real-time temperature, humidity, wind speed and direction to form key information on fire hazards. The edge detection and perception module 100 sends the key information on fire hazards to the communication controller 210.

[0048] (2) Edge-cloud collaborative transmission: The transmission mechanism of the edge-cloud collaborative transmission module 200 is consistent with the "workflow based on illegal intrusion". The communication controller 210 prioritizes the 4G network. After the timeout, it switches to the Beidou short message satellite mode and simplifies the key information of the fire hazard into a message with "timestamp, coordinates, and type of abnormal event" and sends it through the satellite. The data type is "suspected fire". The complete audio clips and video frames are temporarily stored in the local cache module 230 and resumed to the cloud through the TLS 1.3 encrypted channel after the network is restored.

[0049] (3) Cloud-based intelligent decision-making: ①Information aggregation: The multi-source data fusion module 310 receives encrypted data from multiple edge-cloud collaborative transmission modules 200 through a message queue, decrypts the encrypted data to obtain decrypted data, and sends the decrypted data to the multi-scenario anomaly judgment module 320.

[0050] ②Multi-scenario anomaly analysis: Based on the initial anomaly identification results, the cloud-based 300 system automatically invokes the fire hazard model 322 for anomaly assessment using the multi-scenario anomaly analysis module 320. The input data for the fire hazard model 322 includes: Topographic parameters consisting of slope, aspect, and elevation; Combustion parameters, which consist of vegetation type, combustible load, and canopy density; Meteorological parameters consisting of temperature, humidity, wind speed, wind direction, and precipitation.

[0051] Slope, aspect, and altitude are obtained based on BeiDou or GPS positioning satellites; vegetation type is identified based on satellite remote sensing imagery, and combustible load is obtained by inversion from satellite remote sensing imagery; temperature, humidity, wind speed, and wind direction data are obtained based on data collected by environmental sensor 130, and rainfall is obtained based on weather forecast data from meteorological stations.

[0052] The fire hazard model 322 is an LSTM (Long Short-Term Memory) model. It is trained using historical data from the protected area, including topographic, combustible material, and meteorological parameters, as input, and fire spread data after a fire occurs as labels. The fire spread data is calculated based on historical satellite remote sensing data. Based on the current input data, the fire hazard model 322 predicts the fire spread area at different future time periods, determines the fire scale at different future time periods, and provides a comprehensive risk score based on the fire scale within the set future time periods.

[0053] This embodiment provides a comprehensive risk score based on the fire size within the next hour. If the predicted fire size is less than 0.01 hectares after one hour, it indicates no fire spread, with a comprehensive risk score of 0.2. If the predicted fire size is between 0.01 and 0.1 hectares, it indicates a slow fire spread, with a comprehensive risk score of 0.4, and the fire is easy to extinguish. If the predicted fire size is between 0.1 and 1 hectares, it indicates a moderate fire spread, with a comprehensive risk score of 0.6, and the fire may spread to the tree canopy. If the predicted fire size is between 1 and 10 hectares, it indicates a high fire spread, with a comprehensive risk score of 0.8, and obvious crown fire characteristics. If the predicted fire size is greater than 10 hectares, it indicates an extremely high fire spread, with a comprehensive risk score of 1.0, and extremely difficult to extinguish.

[0054] ③ Emergency command generation: When the comprehensive risk score is not less than 0.4, the fire shows a tendency to spread. Emergency response instructions for fire hazards are generated in JSON format based on the comprehensive risk score and the predicted fire spread area.

[0055] ④ Atlas update: Based on the anomaly analysis and risk assessment results from multiple scenarios, the map generation module 320 generates an ecological situation map. This map is updated using a GIS engine, marking the boundaries of predicted fire spread areas for multiple future time periods and displaying the status of event response.

[0056] (4) Emergency dispatch: After receiving the emergency response instruction for fire hazard, the task scheduling and tracking module 410 performs the following operations: Resource matching: A fire fighting plan is determined based on a comprehensive risk score. This plan includes the types and quantities of firefighting equipment, the number of patrol personnel, fire fighting routes, and evacuation areas for affected personnel. The task scheduling and tracking module 410 queries the patrol terminals 420 near the location of a "suspected fire" incident based on the emergency response command for fire hazards, and queries the available drone inspection terminals 120.

[0057] Task Push: The task scheduling and tracking module 410 pushes emergency response instructions for fire hazards to the App on the patrol personnel's patrol terminal 420. The App sends fire fighting and evacuation tasks to the patrol personnel, and the patrol terminal 420's App simultaneously publishes these tasks. Simultaneously, the task scheduling and tracking module 410 issues instructions to multiple drone inspection terminals 120 to proceed to the location of the "suspected fire." The drone inspection terminals 120 are used to patrol near the "suspected fire" location, collecting real-time fire image data.

[0058] Information dissemination: The monitoring screen 430 issues an alarm and displays an ecological situation map. At the same time, the system uses the early warning broadcasting equipment 440 near the evacuation area to repeatedly play voice warnings, explaining the location of the "suspected fire" incident and reminding affected personnel to stay away from the incident area.

[0059] On-site handling and closed-loop tracking: After patrol personnel sign off on the task on the app, the system begins timing and tracking. Once the patrol personnel arrive at the scene with firefighting equipment, they use the app to take and upload photos or short videos of the fire scene to fight the fire. If the overall risk score is not less than 0.6, the task dispatch and tracking module 410 automatically forwards the alarm to the superior fire command center and requests reinforcements.

[0060] The task scheduling and tracking module 410 continuously receives feedback from drones and ground personnel, dynamically adjusts the firefighting plan and evacuation area, and updates the incident handling status on the monitoring screen 430. Once the patrol personnel confirm that the open flames are extinguished and there is no risk of reignition, they submit a "handling completed" report via the App and upload photos of the scene. Based on the feedback, the cloud-based system 300 changes the incident handling status in the situation map from "handling" to "pending review".

[0061] After the incident, the system automatically packages all data to generate an unalterable "Fire Response Retrospective Archive," which includes trigger smoke / temperature data, satellite alarm information, fire situation assessment logs, spread prediction process, dispatch command sequence, drone reconnaissance video, firefighter trajectories, and response results. Relevant personnel can analyze the cause of the fire through the retrospective archive to optimize fire prevention plans and the deployment of edge nodes.

[0062] The workflow based on species anomalies is as follows: (1) Edge detection and sensing: Edge monitoring nodes are deployed in a grid-like manner in the main water sources, rare species nesting / migration corridors, and areas with high incidence of animal diseases over the years within the protected area. Each node consists of an infrared camera 110, an acoustic monitoring device 140, and an optional drone inspection terminal 120. All devices are connected to a lightweight inference module 150.

[0063] Acoustic screening and cleaning: Acoustic monitoring device 140 continuously collects ambient sound at a sampling rate of 16kHz, deployed in the tree canopy or on fixed poles, 3-5 meters above the ground. Data cleaning unit 151 uses spectral subtraction to reduce noise, filtering steady-state background noise such as wind and rain, retaining only audio segments with a signal-to-noise ratio greater than 10dB. When the energy proportion of biological sounds detected exceeds 60%, initial target identification is triggered. After triggering, infrared camera 110 performs differential analysis; if the pixel change rate of multiple consecutive video frames is less than 5%, the video is discarded.

[0064] Initial target identification: The target initial identification unit 152 loads two types of lightweight models for species anomalies in parallel on a low-power AI chip. The lightweight models are as follows: 1) Bird and Mammal Call Classification Model: The preferred model is the MobileNetV3 model, which can extract Mel frequency cepstral coefficient features to identify alarm calls and flock calls of various key indicator species within the protected area. In this embodiment, the number of key indicator species is 40. If alarm calls occur frequently outside the breeding season, or if the intensity of flock calls exceeds three times the historical baseline, it is identified as a "suspected population anomaly" event.

[0065] 2) Animal Target Visual Detection Model: The preferred model is the YOLOv5s model, which can detect video captured by the infrared camera 110 frame by frame and identify large-scale animal deaths and abnormal gatherings. If any of the above situations are detected and the confidence level is greater than 0.65, it is judged as a "suspected species anomaly" event. If the drone inspection terminal 120 captures large-scale animal deaths and abnormal gatherings, it is also judged as a "suspected species anomaly" event.

[0066] Compressed and packaged: After a suspected species anomaly occurs, the compression encoding unit 153 compresses and packages the structured information, such as the alarm sound clip (5 seconds), key video frames, anomaly type, species ID, individual quantity statistics, GPS coordinates and timestamps, into key information about the species anomaly and sends it to the communication controller 210.

[0067] (2) Edge-cloud collaborative transmission: The workflow is consistent with the "workflow based on illegal intrusion". The communication controller 210 prioritizes the 4G network. After the timeout, it switches to the Beidou short message satellite mode and simplifies the key information of the fire hazard into a message with "timestamp, coordinates, and type of abnormal event" and sends it through the satellite. The data type is "suspected species anomaly". The complete audio clips and video frames are temporarily stored in the local cache module 230 and will be resumed to the cloud through the TLS 1.3 encrypted channel after the network is restored.

[0068] (3) Cloud-based intelligent decision-making: ① Information aggregation: The multi-source data fusion module 310 receives and decrypts the key information, obtains the decrypted data, and sends the decrypted data to the multi-scenario anomaly judgment module 320.

[0069] ② Anomaly analysis in multiple scenarios: The cloud-based multi-scenario anomaly analysis module 320 calls the species anomaly model 323. The input data for the species anomaly model 323 includes: Location, species, and on-site statistics of "suspected species anomaly" events; Historical activity patterns of anomalous species populations in this region; Recent abnormal weather records; On-site statistical information includes Mel frequency cepstral coefficient characteristics collected by acoustic detection equipment 140 and video footage captured by infrared camera 110. Historical activity patterns include seasonal migration times and diurnal rhythms, retrieved from historical data. Recent weather anomaly records include droughts, floods, and extreme low temperatures, obtained from the associated meteorological bureau interface or the protected area's own meteorological station database.

[0070] Species anomaly model 323 is a causal Bayesian network. Its topology was jointly defined by ecologists and veterinary experts in the protected area, and it includes nodes such as "changes in food resources," "abnormal weather," "disease outbreaks," "abnormal population aggregation," and "abnormal calls." The conditional probability table was determined in the following way.

[0071] Species anomaly cases are statistically analyzed from a historical anomaly event database to obtain frequencies, which are then converted into probabilities. Species anomaly cases include extreme drought, decreased rodent density, and increased alarm call frequency. During training, population anomaly event data recorded in the reserve over several years are used as the training set. The network parameters are learned through maximum likelihood estimation to form an inference engine capable of causal inference. In this embodiment, the training set is obtained from the reserve's records over the past 5 years.

[0072] Based on the current input, the model infers the root cause of the anomaly, outputs a comprehensive risk score, and generates species anomaly coordinates, which are acquired by the acoustic detection device 140 or the infrared camera 110. When a "suspected species anomaly" event is determined based on a bird and mammal call classification model, the installation location of the acoustic detection device 140 at the event location is used as the species anomaly coordinates; when a "suspected species anomaly" event is determined based on an animal target visual detection model, the installation location of the infrared camera 110 at the event location is used as the species anomaly coordinates.

[0073] If the diagnosis is a highly contagious disease and the species is a core protected species, the score is greater than 0.9; If there is abnormal fluctuation in the population but no deaths have been found yet, the score is between 0.7 and 0.9; If the injury or illness is an isolated incident and the score is less than 0.7, it should be recorded, archived, and continuously monitored.

[0074] ③ Emergency command generation: When the comprehensive risk score is less than 0.7, no emergency response instruction is generated; when the comprehensive risk score is not less than 0.7, a species anomaly emergency response instruction in JSON format is generated based on the species anomaly coordinates and the comprehensive analysis score.

[0075] ④ Atlas update: The map generation module 330 updates the ecological status map on the GIS engine, marks the abnormal locations of species, the habitat range of abnormal species, and displays the handling status of the event.

[0076] (4) Emergency dispatch: After receiving the species anomaly emergency response instruction, the task scheduling and tracking module 410 performs the following operations: Resource matching: Patrol terminals 420 and drone inspection terminals 120 are matched according to risk level. When the comprehensive risk score is greater than 0.9, patrol terminals 420 with patrol personnel having a veterinary background are matched.

[0077] Task push: The task scheduling and tracking module 410 sends verification and sampling tasks through the App of the patrol terminal 420, and dispatches the drone inspection terminal 120 to the location of the "suspected species anomaly" event to collect and transmit images.

[0078] Information dissemination: The monitoring screen 430 issues an alarm and displays an ecological status map; when the comprehensive risk score is greater than 0.9, a warning is broadcast near the location of the "suspected species anomaly" incident via the early warning broadcasting equipment 440 to remind unauthorized personnel to stay away from the area.

[0079] Closed-loop execution: After patrol personnel sign for the package, they rush to the scene, collect samples, take photos and videos as instructed, and report them. After the handling is completed, the cloud changes the event status to "pending review," and the system packages the data to generate an unalterable "Species Anomaly Handling Review Archive."

[0080] The workflow based on habitat destruction is as follows: (1) Edge detection and sensing: Habitat destruction monitoring primarily relies on regular aerial photography by drone inspection terminals (120) and fixed deployment of environmental sensors (130). Drone hangars are located at the management stations of the protected area and the boundaries of key ecological types, conducting monthly or weekly patrols and photography of key areas within the protected area according to preset routes. Environmental sensors (130) are deployed in a grid pattern along major rivers, lakes, reservoirs, and typical vegetation plots. Key areas include ecological red line zones and areas historically prone to deforestation.

[0081] Image cleaning and registration: The data cleaning unit 151 performs radiometric and geometric correction on the high-resolution images transmitted back by the UAV inspection terminal 120, eliminating differences in illumination and shadow, and performs sub-pixel registration with the locally pre-stored historical base maps. The environmental sensor 130 also includes a turbidimeter, a pH sensor, a dissolved oxygen concentration sensor, and a soil moisture sensor to continuously monitor water quality and soil moisture.

[0082] Initial target identification: The target initial identification unit 152 loads the quantized lightweight habitat destruction model, preferably the Unet-CBAM model, and performs semantic segmentation on the registered image pairs to identify the following habitat change types: Decreased vegetation cover: Areas where the NDVI index decreases by less than 0.2; Newly added bare soil or logging sites: patches with a continuous area greater than 100 square meters; Water pollution or shrinkage: abnormal water quality or significant reduction in water area; For "suspected habitat destruction" events such as water pollution or shrinkage, if the water quality parameters measured by the environmental sensor 130 exceed the Class I water standard limit for 30 consecutive minutes, it indicates abnormal water quality. If the water area detected by the drone inspection terminal 120 has shrunk by 20% compared to the previous year, it also indicates abnormal water quality.

[0083] Compressed and packaged: The change detection difference map, habitat change type, area of ​​change zone, coordinates of the center of change zone, water quality parameters, soil environmental parameters, and timestamp are compressed and packaged to form key information on habitat destruction and sent to the communication controller 210. The change detection difference map is a comparison map collected by the UAV inspection terminal 120 in the same area and at the same time of year this year and last year.

[0084] (2) Edge-cloud collaborative transmission: The transmission mechanism is the same as above. The summary of the "suspected habitat destruction" event is sent via Beidou short message satellite. The complete difference map and sensor data are cached locally and will be encrypted and transmitted again after the network is restored. I will not go into too much detail.

[0085] (3) Cloud-based intelligent decision-making: ① Information aggregation: The multi-source data fusion module 310 receives and decrypts key information and sends it to the multi-scenario anomaly analysis module 320.

[0086] ② Multi-scenario Anomaly Assessment: The cloud-based habitat destruction model 324 is invoked to determine the type of habitat change, outputting verification coordinates and a comprehensive risk score. These verification coordinates are the coordinates of the "suspected species anomaly" incident location, obtained by the UAV inspection terminal 120. The model's input data includes: Area of ​​the changed region, coordinates of the center of the changed region, water quality parameters, and soil environmental parameters; The satellite remote sensing time series images of the region over the past three years were retrieved from a historical remote sensing database in the cloud and used to calculate the rate of change and historical trends; The spatial overlay relationship between the damaged area and the habitat of the protected species is obtained by overlaying the coordinates of the changed area with the functional zoning layer and the species distribution layer of the protected area. Soil and topographic data are retrieved from the basic geographic information database of the protected area.

[0087] Methods for constructing and training the habitat destruction model 324: The model is a 3D convolutional network for spatiotemporal variation analysis, fusing time-series remote sensing features and static environmental features. Its structure includes: Temporal encoding branch: A 3D-CNN (such as a cropped version of C3D) is used to extract features from time-series slices of remote sensing images from recent years. 3D-CNN is a three-dimensional convolutional neural network. This embodiment extracts features from time-series slices of remote sensing images from the past three years. The study also learns long-term and short-term patterns of vegetation change. Vegetation change includes natural succession, gradual degradation, and sudden deforestation.

[0088] Spatial context branch: Use the U-Net encoder to extract the spatial correlation between the current change map and static features such as terrain and soil; Fusion and Classification Head: After the two feature sets are concatenated, they pass through a fully connected layer, simultaneously outputting habitat change type (natural disturbance / legal human activity / suspected illegal activity) and a comprehensive risk score (0-1 range). The output habitat change type includes decreased vegetation cover, newly added bare soil or logging areas, and water pollution or shrinkage. Training data uses confirmed habitat destruction events from the protected area's historical archives as labels, and the samples include time-series images and change detection maps of the event locations from the most recent several years. The samples in this embodiment are derived from images and change detection maps from the past 3 years. The comprehensive risk score levels are divided as follows: When damage occurs in areas located within ecological red lines or habitats of critically endangered species, the overall risk score is greater than 0.9. The comprehensive risk score for damage to habitats located outside ecological red line areas and critically endangered species habitats, but falling under one of the three types of habitat change—decreased vegetation cover, newly added bare soil or logging sites, or water pollution or shrinkage—is 0.7-0.9. When the destruction occurs outside of ecological red line areas and habitats of critically endangered species, but is deployed in any of the three types of habitat change mentioned above, the overall risk score is less than 0.7.

[0089] ③ Emergency command generation: The emergency command generation module 340 outputs habitat destruction emergency response commands in JSON format based on comprehensive risk scores, habitat change types, and verification coordinates.

[0090] ④ Atlas update: The map generation module 330 updates the map on the GIS engine, overlays the boundaries of the damaged area, the comprehensive risk score, the habitat range of the affected protected species, and marks the status of the event handling.

[0091] (4) Emergency dispatch: After receiving the emergency response instruction, the task scheduling and tracking module 410 executes the following: Resource matching: Match patrol terminals 420 of patrol personnel near the "suspected habitat destruction" incident, and dispatch drone inspection terminals 120 to collect evidence from the air.

[0092] Task push: The task scheduling and tracking module 410 sends emergency response instructions through the App of the patrol terminal 420, so that the patrol personnel go to the location of the "suspected habitat destruction" incident and the drone inspection terminal 120 takes orthophotos of the incident location.

[0093] Information dissemination: The monitoring screen 430 issues an alarm and displays a map; when the comprehensive risk score is greater than 0.9, the warning broadcast device 440 plays a warning.

[0094] Closed-loop enforcement and restoration tracking: After patrol personnel sign for the equipment, they arrive at the scene to stop the damage, temporarily seize the tools, and report the evidence. After the disposal is completed, the cloud will change the status to "pending review" and package it into an unalterable "Habitat Destruction Disposal Review File" for subsequent restoration assessment and model iteration.

[0095] As can be seen from the detailed description of the above embodiments, the system proposed in this invention, through its innovative layered architecture and the mechanism of close collaboration among each layer, successfully extends AI capabilities to the edge of harsh environments, overcomes network obstacles through intelligent communication methods, and realizes global intelligent decision-making and automated emergency dispatch in the cloud. Ultimately, it forms a complete, efficient, and evolvable solution for ecological monitoring and emergency response in nature reserves, and completely solves the various technical problems pointed out in the background art.

[0096] Second Embodiment like Figure 8 As shown, this embodiment relates to a method for ecological monitoring and emergency response in a border-cloud collaborative nature reserve, including the following steps: Step 1: Collect multi-source ecological monitoring data through the edge terminal 100, and use the lightweight inference module 150 built into the edge terminal 100 to perform data preprocessing and initial anomaly identification to filter out key information; Step 2: Through the edge-cloud collaborative transmission module 200, the communication controller 210 adaptively selects the communication mode according to the current network status and encrypts and uploads the key information identified in Step 1 to the cloud 300. Step 3: The cloud 300 gathers key information from multiple edge terminals 100 through the multi-source data fusion module 310, calls the multi-scenario anomaly judgment module 320 to perform global anomaly judgment and risk level assessment, and constructs an ecological status map of the protected area. The emergency command output module 330 generates standardized emergency response commands based on the global anomaly judgment and risk level assessment results. Step 4: The emergency dispatch execution module 400 receives the emergency response instruction, performs patrol resource scheduling through the task scheduling and tracking module 410, pushes the response task to the patrol terminal 420, and alarms through the monitoring screen 430 and the early warning broadcasting equipment 440 to complete the response to the emergency. Step 5: The emergency dispatch execution module 400 tracks the progress of task handling in real time and transmits the on-site verification data back to the cloud 300, which then updates the ecological situation map and generates a review file.

[0097] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A border-cloud collaborative nature reserve ecological monitoring and emergency response system, characterized in that, include: At the edge, ecological monitoring data of the protected area is collected, preprocessed and initially identified as anomalies are performed on the ecological monitoring data, the type of abnormal event is determined, and key information is obtained and uploaded. The key information includes the type of abnormal event, the coordinates of the event location, and the ecological monitoring data. The edge-cloud collaborative transmission module is used to receive and upload the key information. In the cloud, it is used to determine the judgment model based on the type of abnormal event according to the received key information, conduct global risk analysis and assessment, generate emergency response instructions, and construct the ecological status map of the protected area; The emergency dispatch execution module has a task dispatch and tracking module and multiple patrol terminals. It is used to receive emergency response instructions and send emergency response instructions to the patrol terminals according to the type of abnormal event and the coordinates of the incident location, and to set up alarms and display ecological situation maps.

2. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 1, characterized in that: The edge-cloud collaborative transmission module includes a communication controller, which adopts an adaptive communication protocol. The adaptive communication protocol dynamically switches between wired communication, wireless communication, and satellite communication modes based on the network signal strength within the nature reserve.

3. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 2, characterized in that: The edge-cloud collaborative transmission module also includes a data encryption module, which is used to encrypt and transmit the key information to the cloud.

4. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 1, characterized in that: The edge device includes a monitoring module for collecting ecological monitoring data, which consists of distributed infrared cameras, drone inspection terminals, environmental sensors, and acoustic monitoring equipment.

5. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 4, characterized in that: When the edge device determines that the abnormal event is an illegal intrusion: key information is obtained based on the infrared camera; the cloud calls the illegal intrusion judgment model, generates a reference location, outputs a comprehensive risk score, and generates an illegal intrusion emergency response command based on the comprehensive risk score and the reference location; the emergency dispatch execution module dispatches patrol terminals and drone inspection terminals to search for illegal intruders based on the illegal intrusion emergency response command.

6. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 4, characterized in that: When the edge device determines that the abnormal event is a fire hazard: key information is obtained based on the infrared camera and environmental sensors; the cloud calls the fire hazard assessment model to predict the fire spread area, outputs a comprehensive risk score, and generates a fire hazard emergency response command based on the fire spread area and the comprehensive risk score; the emergency dispatch execution module dispatches patrol terminals and drone inspection terminals to carry out fire fighting and personnel evacuation based on the fire hazard emergency response command.

7. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 4, characterized in that: When the edge device determines that the type of abnormal event is species abnormality: key information is obtained based on infrared cameras, acoustic monitoring equipment and drone inspection terminals; the cloud calls the species abnormality judgment model to generate species abnormality coordinates, outputs a comprehensive risk score, and generates species abnormality emergency response instructions based on the species abnormality coordinates and comprehensive analysis score. The emergency dispatch execution module dispatches patrol terminals and drone inspection terminals based on species anomaly emergency response instructions to carry out sampling and verification tasks at the incident site.

8. The ecological monitoring and emergency response system for the border-cloud collaborative nature reserve according to claim 1, characterized in that: When the edge determines that the type of abnormal event is habitat destruction: key information is obtained based on the drone inspection terminal and environmental sensors; the cloud calls the habitat destruction assessment model to generate verification coordinates and habitat change type, outputs a comprehensive risk score, and generates habitat destruction emergency response instructions based on the verification coordinates, habitat change type and comprehensive risk score. The emergency dispatch execution module dispatches patrol terminals and drone inspection terminals to the verification coordinates to collect evidence based on habitat destruction emergency response commands.

9. A method for ecological monitoring and emergency response in border-cloud collaborative nature reserves, based on the system described in any one of claims 1-8, characterized in that, Includes the following steps: Step 1: The edge device collects multi-source ecological monitoring data, performs data preprocessing and initial anomaly identification, and filters out key information; Step 2: The edge-cloud collaborative transmission module uploads the key information identified in Step 1 to the cloud; Step 3: The cloud gathers key information from multiple edge devices, calls the multi-scenario anomaly analysis module to perform global anomaly analysis and risk level assessment, constructs an ecological status map of the protected area, and generates emergency response instructions; Step 4: The emergency dispatch execution module receives the emergency response instruction, executes patrol resource dispatch, pushes the response task, alarms, and displays the ecological situation map.

10. The method for ecological monitoring and emergency response in the border-cloud collaborative nature reserve according to claim 1, characterized in that, It also includes step five: the emergency dispatch execution module tracks the progress of task handling in real time and transmits the on-site verification data back to the cloud, which then updates the ecological situation map and generates a review file.

Images