An edge-computing-based field monitoring method and system

Abstract

This invention relates to the field of edge computing technology, specifically to a field monitoring method and system based on edge computing, comprising the following steps: constructing a local data pool by associating background with temperature and humidity; generating preliminary matching results by comparing target pixels; updating the background for known species to generate an adaptive pool; extracting the skeleton of unknown species to generate a lightweight topology map; and encapsulating compliant displacement data to generate a monitoring result package. In this invention, by associating temperature, humidity, and time with edge computing nodes to construct a local background feature data pool; dynamically eliminating and generating an adaptive local background feature data pool based on target matching classification; and extracting skeleton node coordinates and connecting edges for suspected unknown species to generate a lightweight topology skeleton map and encapsulating monitoring results that conform to displacement constraints, this invention achieves dynamic adaptive filtering of background interference in complex field environments and proximity analysis of target data, effectively improving the accuracy of monitoring the physical movements and behavioral characteristics of wild animals.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to a field monitoring method and system based on edge computing. Background Technology

[0002] Edge computing technology primarily involves building distributed platforms that integrate computing and storage capabilities closer to the data source to provide localized computing services. Its main purpose is to reduce data transmission latency and bandwidth pressure on central servers to meet the needs of localized data processing. It is widely used in industries such as smart manufacturing, connected vehicles, smart cities, and video security. Wildlife monitoring methods refer to the process of continuously observing and recording the species, population distribution characteristics, activity patterns, and habitats of wild animals in their natural ecological environments. This is mainly used to collect basic ecological information about wild animals to assist in species conservation and ecological research. Existing technologies such as infrared-triggered cameras, drone aerial patrols, radio telemetry, and satellite collar positioning are typically used to collect image data and location coordinates of wild animals.

[0003] Traditional methods for monitoring wild animals in the wild, while capable of collecting images and location coordinates of wild animals through existing technologies such as infrared trigger cameras, drone patrols, and satellite positioning to assist ecological research, suffer from several drawbacks. These include a large number of redundant environmental images that increase the burden on transmission bandwidth, poor adaptability to environmental changes, and difficulty in accurately analyzing the movement characteristics of unknown species, especially given the complex and ever-changing natural environment and limited network communication conditions. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a field monitoring method based on edge computing.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a field monitoring method based on edge computing, comprising the following steps: S1: Obtain natural background image data to determine initial background texture feature data, associate the initial background texture feature data with the obtained initial outdoor environmental temperature data, initial outdoor environmental humidity data and initial equipment running time data to generate outdoor feature binding combination data, and store the outdoor feature binding combination data in the local storage space of the edge computing node to construct a local background feature data pool. S2: Extract moving targets in the field from the field monitoring image frame data, determine pixel distribution feature data based on the moving targets, and compare the pixel distribution feature data with known species feature data to generate preliminary matching results for the field targets; S3: Based on the preliminary matching results of the known species of the field target, extract the current field background texture feature data of the external region of the field moving target, compare the current field background texture feature data with the initial background texture feature data, remove the field feature binding combination data in the local background feature data pool, and generate a field adaptive local background feature data pool. S4: Based on the preliminary matching results of the wild target suspected to be an unknown species, extract the coordinate data of key skeleton nodes of wild animals based on the image grayscale values ​​of the wild moving target, and combine it with the skeleton node connection edge data generated according to the physical connection distance between the coordinate data of key skeleton nodes of wild animals to generate lightweight topological skeleton graph data of wild animals. S5: Based on the lightweight topology skeleton data of the wild animals, extract the positional changes of the coordinate data of the key skeleton nodes of the wild animals to determine the animal joint displacement values, encapsulate the data that meets the biological displacement threshold constraints, and generate the edge computing wild animal field monitoring result package.

[0006] As a further embodiment of the present invention, the local background feature data pool includes a steady-state reference dictionary, an environmental baseline index, and a spatiotemporal alignment mapping table; the preliminary matching results of the field target include a similarity score matrix, candidate species classification labels, and matching confidence evaluation values; the adaptive local background feature data pool in the field includes dynamic change snapshots, redundancy removal logs, and a scene evolution constant library; the lightweight topological skeleton graph data of wild animals includes spatial morphology tensors, limb contour vectors, and structural weight distribution maps; the edge computing wild animal field monitoring result package includes a behavior determination identifier stream, an action semantic encoding set, and a remote reporting compressed file.

[0007] As a further aspect of the present invention, the step of obtaining the local background feature data pool specifically comprises: S111: Obtain the first natural background image data, extract the image color channel data from the first natural background image data, calculate the color channel distribution state parameters based on the image color channel data, extract the color distribution parameters of the corresponding pixels based on the color channel distribution state parameters, calculate the texture distribution state based on the color parameters, and establish the initial background texture feature data. S112: Obtain initial outdoor ambient temperature data, obtain initial outdoor ambient humidity data, obtain initial equipment operating time data, combine initial outdoor ambient temperature data and humidity data to calculate environmental state distribution parameters, associate the initial background texture feature data, environmental state distribution parameters and operating time data, establish attribute binding relationship and generate outdoor feature binding combination data; S113: Obtain the local storage space of the edge computing node, extract the storage capacity parameters in the local storage space of the edge computing node, calculate the storage address space allocation ratio according to the storage capacity parameters, allocate the corresponding storage address blocks according to the storage space allocation ratio, store the field feature binding combination data into the corresponding storage address blocks to generate set items, and establish a local background feature data pool.

[0008] As a further aspect of the present invention, the step of obtaining the preliminary matching result of the field target specifically includes: S211: Collect field monitoring image frame data, extract field moving targets contained in the field monitoring image frame data at time intervals, extract the image boundary coordinate parameters where the field moving targets are located, locate the target image area based on the image boundary coordinate parameters, calculate the spatial distribution parameters based on the color channel status within the target image area and extract the corresponding parameter items of the target, and obtain pixel distribution feature data; S212: Obtain known species feature data, extract reference species distribution parameters from the known species feature data, establish feature comparison benchmarks based on reference species distribution parameters, calculate the cosine parameter of the angle between the pixel distribution feature data and the known species feature data in the vector space using the cosine similarity algorithm, calculate the spatial distance difference state based on the cosine parameter of the angle, and obtain the target cosine similarity value. S213: Obtain the similarity judgment threshold, extract the lower limit parameter of similarity in the similarity threshold, perform a comparison action between the target cosine similarity value and the similarity judgment threshold, determine the species category identifier of the current moving target based on the comparison difference, extract the target classification item in combination with the species category identifier and merge it into the corresponding species set to establish the output parameter judgment form, and generate the preliminary matching result of the field target.

[0009] As a further aspect of the present invention, the step of obtaining the adaptive local background feature data pool in the field specifically includes: S311: Based on the preliminary matching results of the known species of the field target, extract the image pixel channel data of the outer edge region of the field moving target, extract the current field background texture feature data based on the image pixel channel data, call the Euclidean distance algorithm to calculate the Euclidean metric parameter between the current field background texture feature data and the initial background texture feature data, and obtain the background feature Euclidean distance value. S312: For the determination condition that the Euclidean distance value of the background feature is greater than the background feature difference threshold, calculate the change difference between the current outdoor environmental temperature, humidity and time data and the initial environmental parameters respectively, combine the ratio of each change difference to the standard parameter to generate the corresponding environmental attenuation ratio parameter, perform a weighted summation operation according to the ratio parameter and multiply it into the initial weight data of the feature to obtain the natural attenuation feature weight data. S313: Obtain the feature retention threshold, perform a size comparison between the natural decay feature weight data and the feature retention threshold, locate the data item of the local background feature data pool based on the comparison difference, remove the field feature binding combination data whose natural decay feature weight data is less than the feature retention threshold, update the data distribution parameter item remaining in the storage space, and establish the field adaptive local background feature data pool.

[0010] As a further aspect of the present invention, the step of acquiring lightweight topology skeleton diagram data of wild animals specifically includes: S411: Using a graph neural network model, scan the gray values ​​of pixels in the image area of ​​a suspected unknown species line by line, obtain the gray values ​​of adjacent pixels, perform a comparison operation between the gray values ​​and adjacent gray values, extract the coordinate distribution of pixels with gray values ​​greater than adjacent gray values ​​and generate a set of position parameters, and obtain the coordinate data of key skeleton nodes of wild animals. S412: Based on the point distribution of the key skeleton node coordinate data of the wild animal, calculate the straight-line distance parameter according to any two coordinate data, obtain the physical connection distance threshold, perform a comparison operation between the straight-line distance parameter and the threshold, extract coordinate points with a distance less than the threshold and perform connection pairing, establish the topological relationship parameter item between paired points to generate the edge structure mapping vector, and generate skeleton node connection edge data. S413: Obtain the coordinate data of the key skeleton nodes of the wild animal and the edge data of the skeleton nodes, extract the associated parameters, combine the node coordinates to form network structure parameters, map the network structure parameters to the area of ​​a single field monitoring image, fuse the structural state and spatial feature parameters to generate superimposed combined morphological attribute parameters, and establish lightweight topological skeleton map data of wild animals.

[0011] As a further aspect of the present invention, the step of obtaining the edge computing wildlife field monitoring result package specifically includes: S511: Collect the lightweight topological skeleton map data of the wild animals at adjacent time points, extract the coordinate data of key skeleton nodes of wild animals within the map data, calculate the spatial motion difference parameter of the nodes based on the offset of the coordinate data in adjacent maps, extract the displacement change state vector of the motion difference parameter in the two-dimensional plane to generate the coordinate fluctuation difference parameter distribution item, and obtain the animal joint displacement value. S512: Obtain the upper limit threshold and lower limit threshold of wild animal biological displacement, perform a comparison action between the animal joint displacement value and the upper and lower limit thresholds, determine whether the displacement value falls within a reasonable displacement range, output the numerical compliance logic state based on the comparison calculation result, generate biological activity verification attribute parameters to confirm the regularity of the numerical fluctuation range, and obtain the displacement constraint judgment state quantity. S513: For the conditions that satisfy the displacement constraint judgment state quantity, extract the objects whose animal joint displacement values ​​are between the upper and lower limits, package the lightweight topology skeleton data of the wild animal and the associated animal joint displacement values, merge the two into the same data structure container and perform encapsulation encoding operation to generate the edge computing wild animal field monitoring result package.

[0012] An edge computing-based field monitoring system includes: The background feature construction module acquires natural background image data to determine initial background texture feature data, associates the initial background texture feature data with the acquired initial outdoor environmental temperature data, initial outdoor environmental humidity data and initial equipment running time data to generate outdoor feature binding combination data, and stores the outdoor feature binding combination data in the local storage space of the edge computing node to construct a local background feature data pool. The target preliminary matching module extracts moving targets in the field from the field monitoring image frame data, determines pixel distribution feature data based on the moving targets, and compares the pixel distribution feature data with known species feature data to generate preliminary target matching results. The feature pool adaptive module extracts the current field background texture feature data of the external region of the field moving target based on the preliminary matching results of the known species of field target, compares the current field background texture feature data with the initial background texture feature data, removes the field feature binding combination data in the local background feature data pool, and generates a field adaptive local background feature data pool. The skeleton topology extraction module extracts the coordinate data of key skeleton nodes of wild animals based on the image grayscale values ​​of the wild moving target, based on the preliminary matching results of the wild target suspected of being an unknown species. It then combines the skeleton node connection edge data generated based on the physical connection distance between the coordinate data of the key skeleton nodes of wild animals with the skeleton node connection edge data generated based on the physical connection distance between the coordinate data of the key skeleton nodes of wild animals to generate lightweight topological skeleton map data of wild animals. The monitoring result encapsulation module, based on the lightweight topological skeleton map data of the wild animals, extracts the positional changes of the coordinate data of the key skeleton nodes of the wild animals to determine the animal joint displacement values, encapsulates the data that meets the biological displacement threshold constraints, and generates an edge computing wild animal field monitoring result package.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by constructing a local background feature data pool by associating temperature, humidity and time with edge computing nodes, dynamically eliminating and generating an adaptive local background feature data pool in the wild based on target matching and classification, and extracting skeleton node coordinates and connecting edges for suspected unknown species to generate a lightweight topological skeleton diagram and encapsulate monitoring results that conform to displacement constraints, dynamic adaptive filtering of background interference from complex wild environments and proximity analysis of target data are achieved at the data source. This significantly reduces bandwidth pressure caused by invalid image transmission, reduces central server computing latency, and effectively improves the accuracy of monitoring the physical movements and behavioral characteristics of unknown wild animals. Attached Figure Description

[0014] Figure 1 This is a flowchart of the main steps of the present invention; Figure 2 This is a flowchart of the process for obtaining the local background feature data pool in this invention; Figure 3 This is a flowchart of the process for obtaining preliminary target matching results in the field according to the present invention; Figure 4 This is a flowchart of the field adaptive local background feature data pool acquisition process of the present invention; Figure 5 This is a flowchart of the process for acquiring lightweight topology skeleton diagram data for wild animals according to the present invention; Figure 6 This is a flowchart illustrating the process of acquiring wildlife field monitoring results using edge computing according to the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Please see Figure 1 A field monitoring method based on edge computing includes the following steps: S1: Obtain natural background image data to determine initial background texture feature data, associate the initial background texture feature data with the obtained initial field environment temperature data, initial field environment humidity data and equipment initial running time data to generate field feature binding combination data, store the field feature binding combination data in the local storage space of the edge computing node, and build a local background feature data pool. S2: Extract moving targets in the field from the field monitoring image frame data, determine the pixel distribution feature data based on the moving targets, and compare the pixel distribution feature data with the known species feature data to generate preliminary matching results for the field targets; S3: Based on the preliminary matching results of the field target indicating known species, extract the current field background texture feature data of the external area of ​​the field moving target, compare the current field background texture feature data with the initial background texture feature data, and combine the changes of the current field temperature data, current field humidity data, and current environment sampling time data with the initial field temperature data, initial field humidity data, and equipment initial running time data to generate field environment memory data. Remove the field feature binding combination data in the local background feature data pool to generate the field adaptive local background feature data pool; S4: Based on the preliminary matching results of wild targets indicating suspected unknown species, extract the coordinate data of key skeleton nodes of wild animals based on the image grayscale values ​​of wild moving targets, and combine them with the skeleton node connection edge data generated based on the physical connection distance between the coordinate data of key skeleton nodes of wild animals to generate lightweight topological skeleton map data of wild animals. S5: Based on the lightweight topology skeleton data of wild animals, extract the positional changes of the coordinate data of key skeleton nodes of wild animals to determine the animal joint displacement values, and encapsulate the lightweight topology skeleton data of wild animals that meet the biological displacement threshold constraints and the animal joint displacement values ​​to generate the edge computing wild animal field monitoring result package.

[0017] The local background feature data pool includes a steady-state reference dictionary, an environmental baseline index, and a spatiotemporal alignment mapping table; the preliminary matching results of field targets include a similarity score matrix, candidate species classification labels, and matching confidence assessment values; the adaptive local background feature data pool includes dynamic change snapshots, redundancy removal logs, and a scene evolution constant library; the lightweight topological skeleton map data of wild animals includes spatial morphology tensors, limb contour vectors, and structural weight distribution maps; and the edge computing wild animal field monitoring result package includes a behavior determination identifier stream, an action semantic encoding set, and a remote reporting compressed file.

[0018] Please see Figure 2 Step S1 is as follows: S111: Obtain the first natural background image data, extract the image color channel data from the first natural background image data, calculate the color channel distribution state parameters based on the image color channel data, extract the color distribution parameters of the corresponding pixels based on the color channel distribution state parameters, calculate the texture distribution state based on the color parameters, and establish the initial background texture feature data. The system uses an image acquisition sensor array to obtain first natural background image data with a resolution of 1920 x 1080 pixels. This massive amount of image data is directly sent to the local edge computing architecture for preliminary dimensionality reduction analysis to avoid network latency interference with wildlife monitoring. Then, the red, green, and blue color channels are extracted from the first natural background image data. The grayscale values ​​of each pixel in each of the three color channels are extracted. The sum of all grayscale values ​​for each channel is divided by the total number of pixels to obtain the mean values ​​for the red, green, and blue channels, respectively. The sum of the squares of the differences between each grayscale value and its corresponding mean is extracted and averaged to obtain the channel variance data. The mean and variance data are concatenated to obtain the color channel distribution parameters. For example, if the red channel mean is 120, the green channel mean is 150, and the blue channel mean is 100, with variances of 10, 15, and 12 respectively, the concatenated color channel distribution parameters are represented as a one-dimensional vector containing six values. By combining color channel distribution parameters, the corresponding pixels within a 100x100 pixel matrix in the image center region are located, and the color distribution parameters of each pixel within this matrix are extracted. The grayscale difference between the eight adjacent pixels surrounding the corresponding pixel is extracted, and local binary pattern encoding is calculated. All encoded values ​​are used to construct a histogram, from which the texture distribution is calculated. This histogram data is then concatenated with the color channel distribution parameters to establish initial background texture feature data, generating a 256-dimensional feature vector. By combining the concatenation of mean and variance with local encoding, the limitations of single grayscale calculation are avoided. This is crucial for edge computing nodes to perform wildlife monitoring tasks under complex lighting conditions.

[0019] S112: Obtain initial field ambient temperature data, obtain initial field ambient humidity data, obtain initial equipment operating time data, combine initial field ambient temperature data and humidity data to calculate environmental state distribution parameters, associate initial background texture feature data with environmental state distribution parameters and operating time data, establish attribute binding relationships and generate field feature binding combination data; Initial ambient temperature data was acquired using an integrated temperature and humidity sensor array deployed in the wild. This sensor array is a core component of the IoT sensing layer for wildlife monitoring. The temperature data was sampled once per minute, specifically at 25 degrees Celsius. Initial ambient humidity data, at 65%, was also acquired using the same sensor. The initial operating time data of the device was obtained by calling the timing chip inside the edge device; the current operating duration is 120 hours. The ambient temperature data, ambient humidity data, and device operating time data were each divided by their respective preset maximum ranges: temperature (50 degrees Celsius), humidity (100%), and operating time (1000 hours), to obtain their normalized values. The normalized temperature value (0.5), humidity value (0.65), and time value (0.12) were linearly weighted and summed, then multiplied by corresponding weights previously determined using a grid search method, each with a sum of 1: 0.4, 0.4, and 0.2, respectively, resulting in an environmental state distribution parameter with a total sum of 0.484. The initial background texture feature data containing 256-dimensional vectors obtained in step S111 is called. This feature data is then mapped and associated with the calculated environmental state distribution parameters and runtime data using a key-value pair data structure to establish attribute binding relationships. This process encapsulates the aforementioned multi-dimensional data into a single data packet structure, generating combined field feature binding data. Experimental results show that the binding mechanism integrating temperature and humidity with feature vectors significantly improves the edge computing robustness of the wildlife field monitoring system under extreme climate conditions.

[0020] S113: Obtain the local storage space of the edge computing node, extract the storage capacity parameters in the local storage space of the edge computing node, calculate the storage address space allocation ratio based on the storage capacity parameters, allocate the corresponding storage address blocks according to the storage space allocation ratio, store the field feature binding combination data into the corresponding storage address blocks to generate set items, and establish a local background feature data pool. A storage capacity read command is sent to the local solid-state drive of the edge computing node to assess the data retention capability of the wildlife monitoring equipment in offline mode, obtain the local storage space of the edge computing node, and extract the total capacity and available capacity parameters of the local storage space of the edge computing node. The current total hard drive capacity is 500 gigabytes, and the available capacity is 300 gigabytes. Dividing the available capacity by the total capacity, the storage address space allocation ratio is 60%. Based on this 60% storage space allocation ratio, corresponding storage address blocks are divided in the local file system, specifically allocating a contiguous logical block of 180 gigabytes for dedicated storage. The combined data of wildlife features are converted into binary data streams according to the timestamp sequence and stored in the corresponding storage address block to generate set items. For each piece of combined data stored, a record entry with an auto-incrementing index is created in this address block. This storage action continues until the current batch of combined data is written, establishing a local background feature data pool. Relying on this data pool, the edge computing front end can autonomously complete feature retrieval, meeting the requirements of low power consumption and long battery life in wildlife monitoring scenarios.

[0021] Table 1. Local Storage Resource Allocation for Edge Computing Nodes Storage resource name Resource values resource units Total storage capacity 500 Gigabytes Available storage capacity 300 Gigabytes Allocation percentage 60 percentage As shown in Table 1, the method of allocating storage blocks based on capacity proportion can effectively prevent a single type of data from filling up the storage space. This data pool will serve as the basic database for edge nodes to perform local rapid feature comparison, providing underlying support for edge computing resource scheduling for all-weather wildlife field monitoring.

[0022] Please see Figure 3 Step S2 is as follows: S211: Collect field monitoring image frame data, extract field moving targets contained in the field monitoring image frame data at time intervals, extract the image boundary coordinate parameters where the field moving targets are located, locate the target image area based on the image boundary coordinate parameters, calculate the spatial distribution parameters based on the color channel status within the target image area and extract the corresponding parameter items of the target, and obtain pixel distribution feature data; To capture high-frequency dynamic biological behavior characteristics, a high-frequency camera was used to acquire field monitoring image frames at a rate of 30 frames per second. Moving targets in the field were extracted from the monitoring image frames at 15-frame intervals. The pixel difference between the current frame and the previous frame was compared using a frame difference method. Regions with a difference greater than a pixel change threshold were marked as moving targets. The pixel change threshold was set to 30. The image boundary coordinates of the moving target were extracted, obtaining the coordinates of the top-left corner (150 x 150 x 200 y) and the bottom-right corner (350 x 350 x 400 y) of the target's bounding rectangle. Based on these four image boundary coordinates, the target image region was located, and a 200x200 pixel image matrix was obtained. The pixel distribution patterns of the red, green, and blue color channels within the target image region were extracted. The pixel values ​​of each channel were divided into 16 intervals ranging from 0 to 255. The number of pixels in each interval was counted, and this number was divided by the total number of pixels in the region to obtain the normalized frequency. Based on this, the spatial distribution parameters were calculated. The 16 spatial distribution parameters from each of the three channels were sequentially concatenated, and the corresponding parameter items for the target were extracted to obtain a total of 48-dimensional pixel distribution feature data in the form of a one-dimensional array. By extracting the color space distribution histogram, the interference of drastic local changes in illumination intensity on the overall features was eliminated, thereby significantly optimizing the target locking accuracy of the edge computing module in the complex lighting environment of wildlife monitoring.

[0023] S212: Obtain known species feature data, extract reference species distribution parameters from the known species feature data, establish feature comparison benchmarks based on reference species distribution parameters, use cosine similarity algorithm to calculate the cosine parameter of the angle between pixel distribution feature data and known species feature data in vector space, calculate spatial distance difference state based on the cosine parameter of the angle, and obtain target cosine similarity value. The system reads a local wildlife database to obtain known species characteristic data. This data is generated from standardized images of 50 previously collected and labeled wildlife species and is pre-installed on an edge computing device to reduce reliance on cloud communication for wildlife monitoring. It extracts 48-dimensional reference species distribution parameters from the known species characteristic data, ensuring the parameter format matches the target characteristic data. Based on these 48-dimensional reference species distribution parameters, a feature comparison benchmark is established. The 48-dimensional pixel distribution feature data obtained in step S211 is extracted. The cosine similarity calculation logic is invoked, multiplying each of the 48 values ​​in the pixel distribution feature data with the corresponding 48 values ​​in the reference species distribution parameters, and then summing the results to obtain the vector dot product value. The current dot product is 850. The square root of the sum of the squares of the values ​​in both sets of data is taken to obtain the vector magnitude. Assuming the target data magnitude is 30 and the reference data magnitude is 32. Dividing the dot product value of 850 by the product of the two moduli 30 and 32 (960), the cosine parameter of the angle between the pixel distribution feature data and the known species feature data in vector space is calculated, yielding a value of 0.885. Based on this cosine parameter 0.885, subtracting it from 1 gives a value of 0.115. The spatial distance difference is then calculated, resulting in the target cosine similarity value of 0.885. This calculation logic, by converting the angle into a similarity ratio, eliminates the difference in absolute pixel count caused by the distance between the target and the camera, accurately reflecting the essential similarity level of species color, and greatly improving the edge computing species identification accuracy of wildlife field monitoring systems.

[0024] S213: Obtain the similarity judgment threshold, extract the lower limit parameter of similarity in the similarity threshold, perform a comparison action between the target cosine similarity value and the similarity judgment threshold, determine the species category identifier of the current moving target based on the comparison difference, extract the target classification item in combination with the species category identifier and merge it into the corresponding species set to establish the output parameter judgment form, and generate the preliminary matching result of the field target; The system configuration registry is accessed to obtain the similarity threshold, which was derived from 1000 random species image matching experiments and set to 0.85. This threshold represents the basic limit for a high degree of species feature matching and is also the trigger condition for shifting edge computing power towards higher-level inference tasks. This similarity threshold of 0.85 is extracted as the lower limit parameter for similarity. A comparison is performed between the target cosine similarity value (0.885) obtained in step S212 and the similarity threshold of 0.85. The result shows that the target cosine similarity value is greater than the lower limit parameter, with a comparison difference of 0.035. Based on this positive comparison difference, the current moving target is determined to be highly matched with the corresponding known species features, and the current moving target is assigned a known species category identifier, defined as category number 5. Combining this species category identifier 5, the detection time, coordinate data, and target cosine similarity value of the current target are packaged and extracted into a target classification item. This classification item is written into the corresponding species set labeled 5 in the local database, establishing an output parameter judgment form that includes the species label and matching confidence, ultimately generating the preliminary matching result of the field target. This determination will be used to trigger in-depth validation processes for specific species, thereby ensuring that wildlife field monitoring networks output high-confidence data conclusions at the edge.

[0025] Please see Figure 4 Step S3 is as follows: S311: Based on the preliminary matching results of known species in the wild, extract the image pixel channel data of the outer edge region of the moving target in the wild, extract the current wild background texture feature data based on the image pixel channel data, call the Euclidean distance algorithm to calculate the Euclidean metric parameter between the current wild background texture feature data and the initial background texture feature data, and obtain the background feature Euclidean distance value. Based on the preliminary matching results of the previously determined field target containing the known species of category 5, image pixel channel data of the outer edge region of the moving field target, extending 20 pixels outward from the bounding rectangle, is extracted. Within this extended annular region, the grayscale attributes of each pixel are collected, and the current field background texture feature data is extracted according to the local binary mode calculation rules, generating a 256-dimensional current background feature vector. The Euclidean distance calculation logic is invoked, and for the current field background texture feature data vector and the 256-dimensional initial background texture feature data vector established in step S111, the corresponding dimension values ​​are subtracted and squared sequentially, resulting in 256 squared differences. The entire vector differencing process is accelerated in parallel by an edge computing microprocessor array. All 256 squared differences are accumulated, and the sum is 3250. The square root of this sum is then calculated to determine the Euclidean metric between the current field background texture feature data and the initial background texture feature data. The square root is calculated to be 57.0, which is the background feature Euclidean distance value. Euclidean distance directly reflects the absolute spatial distance between two sets of high-dimensional vectors in space. It can accurately quantify the degree of physical background deformation caused by external lighting or vegetation being blown by the wind. This is a necessary prerequisite for eliminating environmental artifacts in wildlife field monitoring scenarios.

[0026] S312: For the judgment condition that the Euclidean distance value of the background feature is greater than the background feature difference threshold, calculate the difference between the current field temperature, humidity and time data and the initial environmental parameters respectively, combine the ratio of each difference to the standard parameter to generate the corresponding environmental attenuation ratio parameter, perform a weighted summation operation based on the ratio parameter and multiply it into the initial feature weight data to obtain the natural attenuation feature weight data. The background feature difference threshold is obtained by calling the local preset parameter table. This threshold is verified by 500 hours of continuous shooting data in the field and is set to 40.0. Based on the judgment condition that the background feature Euclidean distance value of 57.0 obtained in step S311 is greater than the background feature difference threshold of 40.0, the attenuation calculation logic is initiated. The current outdoor ambient temperature is obtained as 28 degrees Celsius, humidity as 70%, and time as 130 hours. These are subtracted from the initial ambient temperature of 25 degrees Celsius, humidity as 65%, and time as 120 hours recorded in step S112, respectively, to calculate a temperature difference of 3 degrees Celsius, a humidity difference of 5%, and a time difference of 10 hours. Dividing the temperature difference of 3 by the standard temperature tolerance of 10, the humidity difference of 5 by the standard humidity tolerance of 20, and the time difference of 10 by the standard time tolerance of 50, respectively, yields environmental attenuation ratio parameters of 0.3, 0.25, and 0.2. The environmental impact multipliers for these three items were set to 0.4, 0.4, and 0.2, respectively. After multiplying each ratio parameter by its environmental impact multiplier and performing a weighted summation, a value of 0.26 was obtained. This coefficient dynamically reflects the drastic changes in the microclimate surrounding the wildlife monitoring base station. Subtracting 0.26 from 1 yielded 0.74. This 0.74 was then multiplied by the initial feature weight data set to 1.0, resulting in a final value of 0.74 for the natural decay feature weight data. By using weighted differences from multi-source environmental data, this process reduces the probability of background features being misjudged due to drastic daily temperature and humidity changes, significantly enhancing the state perception capability of edge computing nodes under variable climate conditions.

[0027] S313: Obtain the feature retention threshold, perform a size comparison between the natural decay feature weight data and the feature retention threshold, locate the data items in the local background feature data pool based on the comparison difference, remove the field feature binding combination data whose natural decay feature weight data is less than the feature retention threshold, update the data distribution parameter items remaining in the storage space, and establish the field adaptive local background feature data pool. The system reads the preset judgment rules from the memory to obtain the feature retention threshold. This threshold, optimized through a 6-month retention test under varying climatic conditions, is ultimately fixed at 0.65, serving as the bottom line for data discard and effectively preventing the continuous erosion of edge computing memory by redundant historical data from wildlife monitoring. A comparison operation is performed between the naturally decaying feature weight data (0.74) obtained in step S312 and the feature retention threshold (0.65), resulting in a difference of positive 0.09. Based on the index markers of the comparison results, each data item within the generated local background feature data pool is located. The weight values ​​of all data items in the data pool are cyclically compared with 0.65, and deletion instructions are executed to remove all wildlife feature binding combinations whose naturally decaying feature weight data is less than the feature retention threshold of 0.65. Data items greater than or equal to this threshold are retained. The data distribution parameter record table remaining in the storage space is updated, the memory fragmentation pointer is reset, and the latest adaptive local background feature data pool for the wild is established.

[0028] Table 2 Update Status Table of Local Background Feature Data Pool Data status attributes Number of features Weighted mean Data volume before update 500 0.62 Data volume to be removed 150 0.45 Updated data volume 350 0.81 As shown in Table 2, after removing inferior data through the threshold comparison mechanism, the average weight of features in the data pool is significantly improved, ensuring the continuous reliability of the background reference benchmark in complex field environments. This enables edge computing terminals to provide a solid and self-cleaning data foundation for persistent wildlife field monitoring.

[0029] Please see Figure 5 Step S4 is as follows: S411: Using a graph neural network model, scan the gray values ​​of pixels in the image area of ​​a suspected unknown species line by line, obtain the gray values ​​of adjacent pixels, perform a comparison operation between the gray values ​​and adjacent gray values, extract the coordinate distribution of pixels with gray values ​​greater than adjacent gray values ​​and generate a set of position parameters, and obtain the coordinate data of key skeleton nodes of wild animals. A graph neural network model, pre-trained with 10,000 labeled animal keypoint images, is loaded. This model includes a feature extraction layer with an input resolution of 224x224 and three graph convolutional layers. After quantization pruning, it is depth-adapted to the lightweight edge computing chip platform used in wildlife field monitoring. The model is then used to sequentially scan the grayscale values ​​of each pixel within the image region of a suspected unknown species. For the currently selected center pixel, the grayscale values ​​of its eight neighboring pixels are obtained. A comparison is performed between the obtained center grayscale value and its eight neighboring grayscale values. The coordinates of pixels whose center grayscale value is strictly greater than all eight neighboring grayscale values ​​are extracted and recorded in a two-dimensional array to generate a set of positional parameters. For example, local extrema with coordinates of 120 (horizontal axis), 150 (vertical axis), 180 (horizontal axis), and 160 (vertical axis) are selected. The set of location parameters is input into the graph convolutional layer of the model. Through feature aggregation, discrete extreme points are clustered into 15 skeletal hinge points to obtain the coordinate data of key skeletal nodes in wild animals. This execution logic, by employing rigorous local maximum comparison combined with graph neural aggregation, avoids pseudo-joint interference caused by animal fur texture, thus improving the anti-animation robustness of the edge computing model in wild animal field monitoring tasks.

[0030] S412: For the point distribution of key skeleton node coordinate data of wild animals, calculate the straight-line distance parameter based on any two coordinate data, obtain the physical connection distance threshold, perform a comparison operation between the straight-line distance parameter and the threshold, extract coordinate points with a distance less than the threshold and perform connection pairing, establish the topological relationship parameter item between paired points to generate the edge structure mapping vector, and generate skeleton node connection edge data. For the output of key skeletal node coordinate data of wild animals containing 15 points, the system sequentially selects any two coordinate data points within the point set. The straight-line distance parameter between the two points is calculated by taking the square root of the sum of the squared differences in their x-coordinates and y-coordinates. For example, if point 1 has x-axis coordinates of 10 and y-axis of 20, and point 2 has x-axis coordinates of 40 and y-axis of 60, the calculated straight-line distance parameter is 50. A physical connection distance threshold is obtained from the local rule base. This threshold is set to 80 pixels, referencing the typical skeletal proportions of medium to large wild animals. This effectively accounts for the impact of target size differences on edge computation space constraints during wild animal monitoring. All calculated straight-line distance parameters are compared with the physical connection distance threshold of 80. Point pairs with a calculated distance greater than or equal to 80 are discarded, while points with a distance less than the threshold are extracted. These two points are then used as valid edge association nodes for connection pairing. A parameter item establishing the topological relationship between paired points is generated, using the start and end coordinates as key-value pairs. One-dimensional tensor quantization is performed on each topological relation parameter to generate edge structure mapping vectors. All mapping vectors are then aggregated to generate skeleton node connection edge data, laying an unambiguous topological foundation for subsequent edge computing behavior intention prediction and full-cycle tracking of wild animals in the wild.

[0031] S413: Obtain the coordinate data of key skeleton nodes of wild animals and the edge data of skeleton nodes, extract the associated parameters, combine the node coordinates to form network structure parameters, map the network structure parameters to the area of ​​a single field monitoring image, fuse the structural state and spatial feature parameters to generate superimposed combined morphological attribute parameters, and establish lightweight topological skeleton map data of wild animals. The process involves acquiring the coordinates of key skeletal nodes in wildlife and their connecting edge data. The association parameters recording the start and end indices of these nodes are extracted from the edge data. The coordinates of 15 nodes are combined according to the pointing order of these association parameters to construct a one-to-one graph Laplacian matrix, thus forming the network structure parameters representing the skeletal connection pattern. These network structure parameters, including the node matrix and edge adjacency matrix, are mapped to the current single field monitoring image area according to their absolute coordinate range. The mean value of the local color channels within a 5x5 pixel square surrounding each node's image area is extracted as a spatial feature parameter. The spatial feature parameter representing the network structure state of topological connections is concatenated with this mean value using vector dimensions, generating a superimposed combination of morphological attribute parameters with location coordinates, connection relationships, and local color. All parameters with multimodal attributes are integrated to form a standardized graph data set, ultimately establishing lightweight topological skeleton graph data for wildlife. This purifies the previously redundant pixel array into a low-dimensional semantic graph structure, highly compatible with the memory-sensitive hardware characteristics of edge computing nodes in wildlife field monitoring systems. By fusing structural parameters with spatial color features, this execution logic significantly improves the spatial representation richness of the skeleton graph when occlusion exists.

[0032] Please see Figure 6 The S5 steps are as follows: S511: Collect lightweight topological skeleton map data of wild animals at adjacent time points, extract the coordinate data of key skeleton nodes of wild animals within the map data, calculate the spatial motion difference parameter of the nodes based on the offset of the coordinate data in adjacent maps, extract the displacement change state vector of the motion difference parameter in the two-dimensional plane to generate the coordinate fluctuation difference parameter distribution item, and obtain the animal joint displacement value. To provide real-time quantitative analysis of the spatiotemporal dynamics of wildlife species, the edge computing module collects lightweight topological skeleton map data of wildlife from two consecutive frames with a timestamp interval of 0.1 seconds via a time synchronization control port. The coordinate data of all 15 key skeleton nodes of wildlife stored within these two image datasets are extracted. For the same skeleton node with the same number, the horizontal offset is obtained by subtracting the horizontal coordinate of the previous frame from the horizontal coordinate of the later frame, and the vertical offset is obtained similarly. Using the offset of the coordinate data within adjacent frames, the arithmetic square root of the sum of the squares of the two offsets is calculated to obtain the spatial motion difference parameter of the corresponding node. This spatial motion difference parameter is extracted and combined with the movement angle calculated from the horizontal and vertical offsets to generate a displacement change state vector in a two-dimensional plane. This process, through localized edge computing processing, effectively ensures the temporal continuity and data privacy of wildlife monitoring in the wild. The displacement change state vectors of all 15 nodes are summarized and arranged to generate a coordinate fluctuation difference parameter distribution item representing the overall skeleton motion characteristics. The difference parameters of the limb distal nodes in this distribution project are extracted and averaged to obtain the animal joint displacement value representing the current movement amplitude. Assuming that the forelimb node moves from horizontal axis 100 and vertical axis 100 to horizontal axis 110 and vertical axis 110, the displacement value is calculated to be 14.1 pixels.

[0033] S512: Obtain the upper and lower limits of animal displacement, perform a comparison between the animal joint displacement value and the upper and lower limits, determine whether the displacement value falls within a reasonable displacement range, output the numerical compliance logic state based on the comparison calculation result, generate biological activity verification attribute parameters to confirm the regularity of the numerical fluctuation range, and obtain the displacement constraint judgment state quantity. The system accesses the local ecological monitoring standard library to obtain the upper limit threshold for wildlife biomigration, representing the normal activity rate, set to 45.0 pixels; simultaneously, it obtains the lower limit threshold for biomigration, excluding environmental noise interference, set to 3.0 pixels. The animal joint displacement value of 14.1 output in step S511 is compared with the aforementioned upper and lower limit thresholds of 45.0 and 3.0. Since 14.1 is greater than 3.0 and less than 45.0, the displacement value is determined to fall within a reasonable displacement range. Based on the comparison result, the system sets the judgment flag to 1 and outputs a verified numerical compliance logic state. Based on this logic state 1, a biomigration verification attribute parameter with a timestamp is generated, confirming that the current numerical fluctuation conforms to the regularity condition of the range of normal animal gait cycles. This provides a highly convincing dynamic basis for the wildlife behavior judgment algorithm in the wild. After the judgment is completed, the Boolean displacement constraint judgment state quantity is output. The current value is true, which strictly limits the reasonable speed range of biological movement, directly filters out the instantaneous movement of artifacts caused by strong wind blowing branches and the small displacement noise caused by camera shaking, and minimizes the useless power consumption of edge computing processor on invalid interference frames.

[0034] S513: For the conditions that satisfy the displacement constraint judgment state quantity, extract the objects whose animal joint displacement values ​​are between the upper and lower limits, package the lightweight topology skeleton data of wild animals and the associated animal joint displacement values, and merge the two into the same data structure container to perform encapsulation and encoding operations to generate the edge computing wild animal field monitoring result package. For the aforementioned output to satisfy the condition of the displacement constraint judgment state quantity, the relevant data of the current target object at that moment, where the animal joint displacement value is between the upper and lower limits, is extracted. This marks the successful complete extraction of a single event of wildlife field monitoring at the edge computing level. The complete lightweight topological skeleton graph data of the wildlife at that moment, along with the associated timestamp, environmental temperature and humidity, and animal joint displacement values, are collected and packaged. The graph structure data and numerical features are then aggregated into a single structure container predefined with a specific data format. A lossless compression algorithm library is called to perform binary encapsulation encoding on all data within the container, converting it into a uniform and compact data packet, generating and outputting the edge computing wildlife field monitoring result package.

[0035] Table 3 Internal Data Structure of Monitoring Results Package Data component name Data capacity value Data format type Skeleton topology data 256 byte array Joint displacement values 32 floating-point format Time and Environment Tags 64 character sequence As shown in Table 3, by uniformly encapsulating and encoding multi-source data, the communication bandwidth occupied by edge computing nodes reporting to the cloud can be minimized, resulting in a significant leap in long-distance transmission efficiency in the field and solving the transmission bottleneck problem of data backhaul in traditional wildlife field monitoring systems. This monitoring result package can be directly used by the cloud server to perform subsequent large-scale species behavior trajectory extrapolation, constructing a cloud-edge collaborative intelligent assessment system for wildlife field monitoring.

[0036] An edge computing-based field monitoring system includes: The background feature construction module is used to execute S1: acquire natural background image data to determine initial background texture feature data, associate the initial background texture feature data with the acquired initial field temperature data, initial field humidity data and initial equipment running time data to generate field feature binding combination data, store the field feature binding combination data in the local storage space of the edge computing node, and build a local background feature data pool. The preliminary target matching module is used to perform S2: extracting moving targets in the field from the field monitoring image frame data, determining pixel distribution feature data based on the moving targets, comparing the pixel distribution feature data with known species feature data, and generating preliminary target matching results; The feature pool adaptive module is used to perform S3: for the preliminary matching results of known species in the field, extract the current field background texture feature data of the external region of the moving target in the field, compare the current field background texture feature data with the initial background texture feature data, remove the field feature binding combination data in the local background feature data pool, and generate the field adaptive local background feature data pool. The skeleton topology extraction module is used to perform S4: based on the preliminary matching results of wild targets suspected to be unknown species, extract the coordinate data of key skeleton nodes of wild animals based on the grayscale values ​​of images of moving targets in the wild, and combine it with the skeleton node connection edge data generated based on the physical connection distance between the coordinate data of key skeleton nodes of wild animals to generate lightweight topological skeleton map data of wild animals. The monitoring result encapsulation module is used to execute S5: based on the lightweight topological skeleton map data of wild animals, extract the positional changes of the coordinate data of key skeleton nodes of wild animals to determine the animal joint displacement values, encapsulate the data that meets the biological displacement threshold constraints, and generate the edge computing wild animal field monitoring result package.

[0037] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A field monitoring method based on edge computing, characterized in that, Includes the following steps: S1: Obtain natural background image data to determine initial background texture feature data, associate the initial background texture feature data with the obtained initial outdoor environmental temperature data, initial outdoor environmental humidity data and initial equipment running time data to generate outdoor feature binding combination data, and store the outdoor feature binding combination data in the local storage space of the edge computing node to construct a local background feature data pool. S2: Extract moving targets in the field from the field monitoring image frame data, determine pixel distribution feature data based on the moving targets, and compare the pixel distribution feature data with known species feature data to generate preliminary matching results for the field targets; S3: Based on the preliminary matching results of the known species of the field target, extract the current field background texture feature data of the external region of the field moving target, compare the current field background texture feature data with the initial background texture feature data, remove the field feature binding combination data in the local background feature data pool, and generate a field adaptive local background feature data pool. S4: Based on the preliminary matching results of the wild target suspected to be an unknown species, extract the coordinate data of key skeleton nodes of wild animals based on the image grayscale values ​​of the wild moving target, and combine it with the skeleton node connection edge data generated according to the physical connection distance between the coordinate data of key skeleton nodes of wild animals to generate lightweight topological skeleton graph data of wild animals. S5: Based on the lightweight topology skeleton data of the wild animals, extract the positional changes of the coordinate data of the key skeleton nodes of the wild animals to determine the animal joint displacement values, encapsulate the data that meets the biological displacement threshold constraints, and generate the edge computing wild animal field monitoring result package.

2. The field monitoring method based on edge computing according to claim 1, characterized in that: The local background feature data pool includes a steady-state reference dictionary, an environmental baseline index, and a spatiotemporal alignment mapping table; the preliminary matching results of field targets include a similarity score matrix, candidate species classification labels, and matching confidence evaluation values; the adaptive local background feature data pool includes dynamic change snapshots, redundancy removal logs, and a scene evolution constant library; the lightweight topological skeleton data of wild animals includes spatial morphology tensors, limb contour vectors, and structural weight distribution maps; the edge computing wild animal field monitoring result package includes a behavior determination identifier stream, an action semantic encoding set, and a remote reporting compressed file.

3. The field monitoring method based on edge computing according to claim 1, characterized in that, The specific steps for obtaining the local background feature data pool are as follows: S111: Obtain the first natural background image data, extract the image color channel data from the first natural background image data, calculate the color channel distribution state parameters based on the image color channel data, extract the color distribution parameters of the corresponding pixels based on the color channel distribution state parameters, calculate the texture distribution state based on the color parameters, and establish the initial background texture feature data. S112: Obtain initial outdoor ambient temperature data, obtain initial outdoor ambient humidity data, obtain initial equipment operating time data, combine initial outdoor ambient temperature data and humidity data to calculate environmental state distribution parameters, associate the initial background texture feature data, environmental state distribution parameters and operating time data, establish attribute binding relationship and generate outdoor feature binding combination data; S113: Obtain the local storage space of the edge computing node, extract the storage capacity parameters in the local storage space of the edge computing node, calculate the storage address space allocation ratio according to the storage capacity parameters, allocate the corresponding storage address blocks according to the storage space allocation ratio, store the field feature binding combination data into the corresponding storage address blocks to generate set items, and establish a local background feature data pool.

4. The field monitoring method based on edge computing according to claim 1, characterized in that, The specific steps for obtaining the preliminary matching results of field targets are as follows: S211: Collect field monitoring image frame data, extract field moving targets contained in the field monitoring image frame data at time intervals, extract the image boundary coordinate parameters where the field moving targets are located, locate the target image area based on the image boundary coordinate parameters, calculate the spatial distribution parameters based on the color channel status within the target image area and extract the corresponding parameter items of the target, and obtain pixel distribution feature data; S212: Obtain known species feature data, extract reference species distribution parameters from the known species feature data, establish feature comparison benchmarks based on reference species distribution parameters, calculate the cosine parameter of the angle between the pixel distribution feature data and the known species feature data in the vector space using the cosine similarity algorithm, calculate the spatial distance difference state based on the cosine parameter of the angle, and obtain the target cosine similarity value. S213: Obtain the similarity judgment threshold, extract the lower limit parameter of similarity in the similarity threshold, perform a comparison action between the target cosine similarity value and the similarity judgment threshold, determine the species category identifier of the current moving target based on the comparison difference, extract the target classification item in combination with the species category identifier and merge it into the corresponding species set to establish the output parameter judgment form, and generate the preliminary matching result of the field target.

5. The field monitoring method based on edge computing according to claim 1, characterized in that, The specific steps for obtaining the adaptive local background feature data pool in the field are as follows: S311: Based on the preliminary matching results of the known species of the field target, extract the image pixel channel data of the outer edge region of the field moving target, extract the current field background texture feature data based on the image pixel channel data, call the Euclidean distance algorithm to calculate the Euclidean metric parameter between the current field background texture feature data and the initial background texture feature data, and obtain the background feature Euclidean distance value. S312: For the determination condition that the Euclidean distance value of the background feature is greater than the background feature difference threshold, calculate the change difference between the current outdoor environmental temperature, humidity and time data and the initial environmental parameters respectively, combine the ratio of each change difference to the standard parameter to generate the corresponding environmental attenuation ratio parameter, perform a weighted summation operation according to the ratio parameter and multiply it into the initial weight data of the feature to obtain the natural attenuation feature weight data. S313: Obtain the feature retention threshold, perform a size comparison between the natural decay feature weight data and the feature retention threshold, locate the data item of the local background feature data pool based on the comparison difference, remove the field feature binding combination data whose natural decay feature weight data is less than the feature retention threshold, update the data distribution parameter item remaining in the storage space, and establish the field adaptive local background feature data pool.

6. The field monitoring method based on edge computing according to claim 1, characterized in that, The specific steps for obtaining the lightweight topology skeleton diagram data of the wild animals are as follows: S411: Using a graph neural network model, scan the gray values ​​of pixels in the image area of ​​a suspected unknown species line by line, obtain the gray values ​​of adjacent pixels, perform a comparison operation between the gray values ​​and adjacent gray values, extract the coordinate distribution of pixels with gray values ​​greater than adjacent gray values ​​and generate a set of position parameters, and obtain the coordinate data of key skeleton nodes of wild animals. S412: Based on the point distribution of the key skeleton node coordinate data of the wild animal, calculate the straight-line distance parameter according to any two coordinate data, obtain the physical connection distance threshold, perform a comparison operation between the straight-line distance parameter and the threshold, establish the topological relationship parameter item between paired points, generate the edge structure mapping vector, and generate skeleton node connection edge data. S413: Obtain the coordinate data of the key skeleton nodes of the wild animal and the edge data of the skeleton nodes, extract the associated parameters, combine the node coordinates to form network structure parameters, map the network structure parameters to the area of ​​a single field monitoring image, fuse the structural state and spatial feature parameters to generate superimposed combined morphological attribute parameters, and establish lightweight topological skeleton map data of wild animals.

7. The field monitoring method based on edge computing according to claim 1, characterized in that, The specific steps for obtaining the edge computing wildlife field monitoring result package are as follows: S511: Collect the lightweight topological skeleton map data of the wild animals at adjacent time points, extract the coordinate data of key skeleton nodes of wild animals within the map data, calculate the spatial motion difference parameter of the nodes based on the offset of the coordinate data in adjacent maps, extract the displacement change state vector of the motion difference parameter in the two-dimensional plane to generate the coordinate fluctuation difference parameter distribution item, and obtain the animal joint displacement value. S512: Obtain the upper limit threshold and lower limit threshold of wild animal biological displacement, perform a comparison action between the animal joint displacement value and the upper and lower limit thresholds, determine whether the displacement value falls within a reasonable displacement range, output the numerical compliance logic state based on the comparison calculation result, generate biological activity verification attribute parameters to confirm the regularity of the numerical fluctuation range, and obtain the displacement constraint judgment state quantity. S513: For the conditions that satisfy the displacement constraint judgment state quantity, extract the objects whose animal joint displacement values ​​are between the upper and lower limits, package the lightweight topology skeleton data of the wild animal and the associated animal joint displacement values, merge the two into the same data structure container and perform encapsulation encoding operation to generate the edge computing wild animal field monitoring result package.

8. The field monitoring method based on edge computing according to claim 6, characterized in that: During the comparison operation, coordinate points with a distance less than a threshold are extracted and connected for pairing.

9. A field monitoring system based on edge computing, characterized in that, The system is used to implement the method according to any one of claims 1-8, comprising: The background feature construction module acquires natural background image data to determine initial background texture feature data, associates the initial background texture feature data with the acquired initial outdoor environmental temperature data, initial outdoor environmental humidity data and initial equipment running time data to generate outdoor feature binding combination data, and stores the outdoor feature binding combination data in the local storage space of the edge computing node to construct a local background feature data pool. The target preliminary matching module extracts moving targets in the field from the field monitoring image frame data, determines pixel distribution feature data based on the moving targets, and compares the pixel distribution feature data with known species feature data to generate preliminary target matching results. The feature pool adaptive module extracts the current field background texture feature data of the external region of the field moving target based on the preliminary matching results of the known species of field target, compares the current field background texture feature data with the initial background texture feature data, removes the field feature binding combination data in the local background feature data pool, and generates a field adaptive local background feature data pool. The skeleton topology extraction module extracts the coordinate data of key skeleton nodes of wild animals based on the image grayscale values ​​of the wild moving target, based on the preliminary matching results of the wild target suspected of being an unknown species. It then combines the skeleton node connection edge data generated based on the physical connection distance between the coordinate data of the key skeleton nodes of wild animals with the skeleton node connection edge data generated based on the physical connection distance between the coordinate data of the key skeleton nodes of wild animals to generate lightweight topological skeleton map data of wild animals. The monitoring result encapsulation module, based on the lightweight topological skeleton map data of the wild animals, extracts the positional changes of the coordinate data of the key skeleton nodes of the wild animals to determine the animal joint displacement values, encapsulates the data that meets the biological displacement threshold constraints, and generates an edge computing wild animal field monitoring result package.

Images