An ai vision-based wild milu deer identification and colony footprint spatiotemporal analysis method

Abstract

The application discloses a kind of based on AI vision's wild Père David's deer identification and community footprint space-time analysis method, comprising the following steps: collecting continuous video data and ground image data, generating individual space-time trajectory sequence;Extract footprint spatial coordinate information, generate footprint space-time distribution dataset;Establish individual-footprint association matrix, generate community footprint association result;Construct space-time discrete grid structure, generate community footprint probability density field;Construct space-time diffusion energy field and generate space-time diffusion energy distribution matrix;Carry out regional division, generate steady-state activity domain and diffusion activity domain;Generate energy accessible path set, calculate structure deviation degree, generate community structure stability determination result;According to community structure stability determination result generates community footprint space-time evolution structure result.The application is based on AI vision and space-time energy modeling, realizes community footprint structure determination and evolution analysis, with the advantages of accurate determination, strong structure interpretability.

Description

Technical Field

[0001] This invention relates to the field of identification and spatiotemporal analysis of footprints, and more particularly to a method for identifying wild elk and spatiotemporal analysis of their footprints based on AI vision. Background Technology

[0002] With the development of ecological monitoring technology, video image-based wildlife identification and activity analysis methods are gradually being applied to the field of group behavior research. Existing technologies typically analyze the activity range of individual animals through target detection, trajectory tracking, or spatial density statistics, and combine hotspot area identification methods to visualize the group activity area. Some methods also introduce time series statistical models to analyze activity frequency, but overall, trajectory distribution or spatial density changes are still the main basis for analysis.

[0003] Existing technologies generally focus on surface trajectory statistics or spatial clustering results, lacking technical means to quantitatively characterize the structural stability and diffusion dynamics corresponding to the spatiotemporal distribution of group footprints. They cannot distinguish the dynamic differences between steady-state active structures and diffusion evolution structures, nor can they systematically determine the stability of group structures, resulting in insufficient accuracy in analyzing group evolution trends and structural change patterns. Summary of the Invention

[0004] One objective of this invention is to propose a method for identifying wild elk and spatiotemporal analysis of their footprints based on AI vision. This invention, based on AI vision and spatiotemporal energy modeling, enables the determination and evolutionary analysis of the footprint structure, and has the advantages of accurate determination and strong interpretability of the structure.

[0005] A method for identifying wild elk and spatiotemporal analyzing their footprints based on AI vision, according to an embodiment of the present invention, includes the following steps: Collect continuous video data and surface image data within the monitoring area, perform target detection, instance segmentation and individual weight recognition on the continuous video data, extract individual identity information, spatial location coordinates and timestamp information, and generate individual spatiotemporal trajectory sequences; Footprint segmentation and 3D morphology reconstruction are performed on surface image data to extract spatial coordinate information of footprints and generate a spatiotemporal distribution dataset of footprints. An individual-footprint association matrix is ​​established based on individual spatiotemporal trajectory sequences and footprint spatiotemporal distribution datasets to generate community footprint association results; A spatiotemporal discrete grid structure is constructed, and the community footprint association results are mapped to the spatiotemporal discrete grid structure. The footprint frequency and individual activity frequency of each grid cell are counted to generate a community footprint probability density field. The energy values ​​of each grid cell are calculated based on the probability density field of the community footprint, and a spatiotemporal diffusion energy field is constructed and a spatiotemporal diffusion energy distribution matrix is ​​generated. Based on the energy threshold, the spatiotemporal diffusion energy distribution matrix is ​​divided into regions to generate steady-state active domains and diffusion active domains; Under the constraints of steady-state and diffusion activity domains, a set of energy-accessible paths is generated. The degree of structural deviation of individual spatiotemporal trajectory sequences in the set of energy-accessible paths is calculated, and the community structure stability determination result is generated. Based on the community structure stability assessment results, the spatiotemporal evolution structure of the community footprint is generated, including the spatial distribution of the community activity core area, the direction of diffusion trend, and the temporal evolution of structural stability.

[0006] Optionally, the generation of the footprint spatiotemporal distribution dataset specifically includes: The surface image data is subjected to image denoising, distortion correction and illumination equalization to generate preprocessed surface image data. Footprint region segmentation is performed on the preprocessed surface image data to generate footprint segmentation results; The footprint segmentation results are divided into connected components to generate a set of footprint regions; Calculate the planar coordinates of each footprint region in the footprint region set to generate footprint planar coordinate data; Three-dimensional morphological reconstruction of the footprint region set is performed based on surface image data to generate three-dimensional footprint morphological data. Morphological features are extracted from the three-dimensional morphological data of footprints to generate footprint morphological feature data; The footprint planar coordinate data, footprint three-dimensional morphological data, and footprint morphological feature data are time-stamped and uniformly organized according to the collection time to generate a footprint spatiotemporal distribution dataset.

[0007] Optionally, the generation of the community footprint association results specifically includes: The individual spatiotemporal trajectory sequences are sorted by time and standardized by spatial coordinates to form a set of individual trajectories containing individual identity information, spatial location coordinates and corresponding timestamp information; The footprint spatiotemporal distribution dataset is processed by coordinate unification and time stamping to form a footprint set containing footprint spatial coordinates and corresponding collection times; For each individual identifier in the individual trajectory set, the spatial coordinates of the footprints in the spatially adjacent area are retrieved within the corresponding timestamp range. The spatial proximity relationship between the individual's spatial location and the footprint's spatial location is calculated, and the time synchronization relationship between the individual's timestamp and the footprint collection time is calculated. When the spatial proximity relationship meets the preset spatial association condition and the temporal synchronization relationship meets the preset temporal association condition, it is determined that there is an association between the individual's identity and the corresponding footprint. All individuals and footprints that meet the association conditions are labeled, and an individual-footprint association matrix is ​​constructed. The rows of the individual-footprint association matrix correspond to the individual identity identifier, the columns correspond to the footprint number, and the matrix elements represent the association status between the corresponding individual and the footprint. The individual-footprint association matrix is ​​cumulatively statistically analyzed within a continuous time window, and individual-footprint pairs that continuously meet the association conditions within multiple continuous time windows are selected to form a stable association set. Community footprint association results are generated based on stable association sets, including the footprint sets corresponding to each individual's identity and their temporal distribution information.

[0008] Optionally, the generation of the community footprint probability density field specifically includes: Under a unified spatial coordinate system, the monitoring area is discretized into grids, and multiple spatial grid units are divided according to a preset spatial resolution. Each spatial grid unit corresponds to a unique spatial range and has a definite grid number. The timeline is divided according to a preset time window length to form a continuous time window sequence; Within each time window, the community footprint association results are mapped to spatial grid cells, the number of footprints in each spatial grid cell is counted, and footprint frequency data is generated. Within the same time window, the number of individual activities in each spatial grid unit is counted based on the community footprint association results, and individual activity frequency data is generated. For each spatial grid cell, within the corresponding time window, the footprint frequency data and individual activity frequency data are weighted according to preset weights to generate a joint frequency value; The joint frequency values ​​of all spatial grid cells within the same time window are normalized to generate the grid probability density values ​​for the corresponding time window. The grid probability density values ​​of each time window are combined in chronological order to generate a community footprint probability density field.

[0009] Optionally, the generation of the spatiotemporal diffusion energy distribution matrix specifically includes: Under the framework of unified spatial grid cells and continuous time windows, the probability density field of community footprints is extracted to obtain the grid probability density value of each spatial grid cell within each time window. For each spatial grid cell, the change in grid probability density value between adjacent time windows is calculated to generate time variation intensity data; For each spatial grid cell, the grid probability density difference between it and its adjacent spatial grid cells is calculated within the same time window to generate spatial diffusion intensity data; In each spatial grid cell and time window, the grid probability density value, time variation intensity data, and spatial diffusion intensity data are combined according to preset weights to generate the corresponding grid energy value. The grid energy values ​​are organized according to the spatial grid cells and time windows to construct a spatiotemporal diffusion energy field; The spatiotemporal diffusion energy field is arranged in a matrix to form a spatiotemporal diffusion energy distribution matrix.

[0010] Optionally, the generation of the steady-state active domain and the diffusion active domain specifically includes: Extract the grid energy values ​​corresponding to each time window in the spatiotemporal diffusion energy distribution matrix; Within each time window, the grid energy values ​​of all spatial grid cells are statistically analyzed, and the average energy value within that time window is calculated. Within the same time window, the squared difference between the grid energy value and the average energy value of all spatial grid cells is calculated, and the squared differences are summed to obtain the energy fluctuation value. The energy threshold for the corresponding time window is determined based on the average energy value and the energy fluctuation value. Within the time window, the grid energy value of each spatial grid cell is compared with the energy threshold. When the grid energy value is less than or equal to the energy threshold, the spatial grid cell is divided into a steady-state active domain. When the grid energy value is greater than the energy threshold, the spatial grid cell is divided into a diffusion active domain. Within the time window, connectivity is determined for spatially adjacent spatial grid cells of the same type, forming a set of stable active domain connected regions and a set of diffuse active domain connected regions, respectively. The sets of connected regions of steady-state activity domains and the sets of connected regions of diffusion activity domains corresponding to each time window are combined in chronological order to form the community structure domain partitioning results.

[0011] Optionally, the generation of the community structure stability determination result specifically includes: Extract the set of spatial grid cells corresponding to the diffusion active domain and the grid energy values ​​in the spatiotemporal diffusion energy field to generate a diffusion active domain grid energy dataset. Within the same time window, a spatially connected network structure is established using spatial grid cells in the diffusion activity domain grid energy dataset as nodes, generating a set of spatially connected nodes and a set of connected edges; For any pair of adjacent nodes in the set of spatially connected nodes, calculate the difference in their corresponding grid energy values ​​to generate a set of path energy costs. In the spatial connectivity network structure, with the path energy cost set as a constraint, starting from the spatial grid cells in the diffusion activity domain, the connected nodes with the minimum path energy cost are selected step by step to generate the energy reachable path set; For each energy-reachable path in the set of energy-reachable paths, the sequence of spatial grid cells it contains and the cumulative energy value of the corresponding grid cells are statistically analyzed to generate path cumulative energy data; The individual spatiotemporal trajectory sequence is mapped to a set of spatial grid cells to generate a set of the individual's actual activity paths; The overlap statistics of the set of actual activity paths of individuals and the set of energy-accessible paths are statistically analyzed at the spatial grid unit level to generate path overlap ratio data; Within the same time window, the cumulative energy data of the path and the path overlap ratio data are jointly calculated to generate a joint path determination value; The path joint determination value is compared with the joint determination threshold to generate the community structure stability determination result.

[0012] Optionally, the generation of the path overlap ratio data specifically includes: Extract the spatial grid cell sequence corresponding to each individual within each time window from the set of actual activity paths of individuals, and generate a set of individual path grid sequence; Extract the spatial grid cell sequence of each energy reachable path in the energy reachable path set within the corresponding time window, and generate a reachable path grid sequence set; Within the same time window, for each individual path in the individual path grid sequence set, a spatial grid cell matching and statistical analysis is performed with each energy-reachable path in the reachable path grid sequence set to generate data on the number of overlapping grid cells. The ratio of the number of overlapping grid cells along a path to the length of the corresponding individual path grid sequence is calculated to generate a single path overlap ratio value. Within the same time window, the overlap ratio of all single paths is statistically analyzed to generate a set of overlap ratios for the time window. The overlap ratios of each time window are arranged in chronological order to form path overlap ratio data.

[0013] Optionally, the generation of the spatiotemporal evolution structure results of the community footprint specifically includes: Extract the community structure stability determination results and the set of connected regions of steady-state activity domains and the set of connected regions of diffusion activity domains to form the structure domain dataset corresponding to the time window; Within each time window, coordinate aggregation is performed on the spatial grid cells in the set of connected regions of the steady-state activity domain to generate spatial distribution data of the community activity core area. Between adjacent time windows, the centroid coordinates of the spatial grid cells in the set of connected regions of the diffusion activity domain are calculated, and the difference between the centroid coordinates of adjacent time windows is calculated to generate diffusion trend direction data. The results of the community structure stability assessment are arranged in chronological order to generate a time series of structure stability. By combining the spatial distribution data of the community activity core area, the diffusion trend direction data, and the time series of structural stability according to the time window, the spatiotemporal evolution structure results of the community footprint are generated.

[0014] The beneficial effects of this invention are: This invention proposes an AI-based vision-based method for identifying wild elk and spatiotemporal analyzing their footprints. It establishes an individual-footprint correlation matrix by constructing a link between individual spatiotemporal trajectory sequences and footprint spatiotemporal distribution datasets. Based on this, it introduces a spatiotemporal discrete grid structure, a community footprint probability density field, and a spatiotemporal diffusion energy field. Furthermore, by dividing the steady-state activity domain into a diffusion activity domain, it establishes a set of energy-accessible paths. Combining the cumulative energy of the paths with the path overlap ratio generates a community structure stability assessment result, ultimately outputting the spatiotemporal evolution structure of the community footprints. Compared to existing methods that rely solely on trajectory statistics or spatial density clustering, this invention goes beyond superficial descriptions of activity frequency or spatial distribution. Instead, it transforms the spatial distribution changes of group activities into quantifiable energy and path structures, characterizing the diffusion dynamics and stability state of the community at the structural level, achieving a technological leap from "statistical description" to "structural assessment."

[0015] Through this technical solution, the present invention can identify the formation and migration trends of the core area of ​​community activity within a continuous time window, distinguish between steady-state activity structures and diffusion evolution structures, and quantitatively analyze the stability of community structures, thereby improving the ability to analyze the laws of population evolution. This method realizes the modeling of the coupling relationship between individual behavioral paths and the dynamics of population structure, giving the community evolution results a clear structural basis and traceability, improving the accuracy and interpretability of the analysis results, and providing a structured, quantifiable, stable and reliable technical means for long-term monitoring and ecological behavior research of wildlife communities. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for identifying wild elk and analyzing their footprints in a community based on AI vision, as proposed in this invention. Figure 2 This is a schematic diagram of energy path construction for a method for identifying wild elk and analyzing their footprints in a community based on AI vision, as proposed in this invention. Figure 3This is a schematic diagram illustrating the structural stability determination of a method for identifying wild elk and analyzing their footprints in a community based on AI vision, as proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A method for identifying wild elk and spatiotemporal analyzing their footprints based on AI vision includes the following steps: Collect continuous video data and surface image data within the monitoring area, perform target detection, instance segmentation and individual weight recognition on the continuous video data, extract individual identity information, spatial location coordinates and timestamp information, and generate individual spatiotemporal trajectory sequences; Footprint segmentation and 3D morphology reconstruction are performed on surface image data to extract spatial coordinate information of footprints and generate a spatiotemporal distribution dataset of footprints. An individual-footprint association matrix is ​​established based on individual spatiotemporal trajectory sequences and footprint spatiotemporal distribution datasets to generate community footprint association results; A spatiotemporal discrete grid structure is constructed, and the community footprint association results are mapped to the spatiotemporal discrete grid structure. The footprint frequency and individual activity frequency of each grid cell are counted to generate a community footprint probability density field. The energy values ​​of each grid cell are calculated based on the probability density field of the community footprint, and a spatiotemporal diffusion energy field is constructed and a spatiotemporal diffusion energy distribution matrix is ​​generated. Based on the energy threshold, the spatiotemporal diffusion energy distribution matrix is ​​divided into regions to generate steady-state active domains and diffusion active domains; Under the constraints of steady-state and diffusion activity domains, a set of energy-accessible paths is generated. The degree of structural deviation of individual spatiotemporal trajectory sequences in the set of energy-accessible paths is calculated, and the community structure stability determination result is generated. Based on the community structure stability assessment results, the spatiotemporal evolution structure of the community footprint is generated, including the spatial distribution of the community activity core area, the direction of diffusion trend, and the temporal evolution of structural stability.

[0019] In this embodiment, the generation of the footprint spatiotemporal distribution dataset specifically includes: The surface image data is subjected to image denoising, distortion correction and illumination equalization to generate preprocessed surface image data. Footprint region segmentation is performed on the preprocessed surface image data to generate footprint segmentation results; The footprint segmentation results are divided into connected components to generate a set of footprint regions; Calculate the planar coordinates of each footprint region in the footprint region set to generate footprint planar coordinate data; Three-dimensional morphological reconstruction of the footprint region set is performed based on surface image data to generate three-dimensional footprint morphological data. Morphological features are extracted from the three-dimensional morphological data of footprints to generate footprint morphological feature data; The footprint planar coordinate data, footprint three-dimensional morphological data, and footprint morphological feature data are time-stamped and uniformly organized according to the collection time to generate a footprint spatiotemporal distribution dataset.

[0020] In this embodiment, the generation of community footprint association results specifically includes: The individual spatiotemporal trajectory sequences are sorted by time and standardized by spatial coordinates to form a set of individual trajectories containing individual identity information, spatial location coordinates and corresponding timestamp information; The footprint spatiotemporal distribution dataset is processed by coordinate unification and time stamping to form a footprint set containing footprint spatial coordinates and corresponding collection times; For each individual identifier in the individual trajectory set, the spatial coordinates of the footprints in the spatially adjacent area are retrieved within the corresponding timestamp range. The spatial proximity relationship between the individual's spatial location and the footprint's spatial location is calculated, and the time synchronization relationship between the individual's timestamp and the footprint collection time is calculated. When the spatial proximity relationship meets the preset spatial association condition and the temporal synchronization relationship meets the preset temporal association condition, it is determined that there is an association between the individual's identity and the corresponding footprint. All individuals and footprints that meet the association conditions are labeled, and an individual-footprint association matrix is ​​constructed. The rows of the individual-footprint association matrix correspond to the individual identity identifier, the columns correspond to the footprint number, and the matrix elements represent the association status between the corresponding individual and the footprint. The individual-footprint association matrix is ​​cumulatively statistically analyzed within a continuous time window, and individual-footprint pairs that continuously meet the association conditions within multiple continuous time windows are selected to form a stable association set. Community footprint association results are generated based on stable association sets, including the footprint sets corresponding to each individual's identity and their temporal distribution information.

[0021] In this embodiment, the generation of the community footprint probability density field specifically includes: Under a unified spatial coordinate system, the monitoring area is discretized into grids, and multiple spatial grid units are divided according to a preset spatial resolution. Each spatial grid unit corresponds to a unique spatial range and has a definite grid number. The timeline is divided according to a preset time window length to form a continuous time window sequence; Within each time window, the community footprint association results are mapped to spatial grid cells, the number of footprints in each spatial grid cell is counted, and footprint frequency data is generated. Within the same time window, the number of individual activities in each spatial grid unit is counted based on the community footprint association results, and individual activity frequency data is generated. For each spatial grid cell, within the corresponding time window, the footprint frequency data and individual activity frequency data are weighted according to preset weights to generate a joint frequency value; The joint frequency values ​​of all spatial grid cells within the same time window are normalized to generate the grid probability density values ​​for the corresponding time window. The grid probability density values ​​of each time window are combined in chronological order to generate a community footprint probability density field.

[0022] In this embodiment, the generation of the spatiotemporal diffusion energy distribution matrix specifically includes: Under the framework of unified spatial grid cells and continuous time windows, the probability density field of community footprints is extracted to obtain the grid probability density value of each spatial grid cell within each time window. For each spatial grid cell, the change in grid probability density value between adjacent time windows is calculated to generate time variation intensity data; For each spatial grid cell, the grid probability density difference between it and its adjacent spatial grid cells is calculated within the same time window to generate spatial diffusion intensity data; In each spatial grid cell and time window, the grid probability density value, time variation intensity data, and spatial diffusion intensity data are combined according to preset weights to generate the corresponding grid energy value. The generation of grid energy values ​​specifically includes: Within each time window, the grid probability density value, temporal variation intensity data, and spatial diffusion intensity data of the corresponding spatial grid cell are extracted to form grid feature combination data. The grid feature combination data undergoes scale unification processing to ensure that the grid probability density value, temporal variation intensity data, and spatial diffusion intensity data are within the same numerical range, generating standardized grid feature data. The standardized grid feature data is then weighted and summed according to preset weight coefficients to generate the grid basic energy value. Within the same time window, the grid basic energy values ​​of all spatial grid cells are normalized to generate the grid relative energy value. Finally, the grid relative energy values ​​of the same spatial grid cell in adjacent time windows are recursively superimposed to generate the final grid energy value. The grid energy values ​​are organized according to the spatial grid cells and time windows to construct a spatiotemporal diffusion energy field; The generation of the spatiotemporal diffusion energy field specifically includes: Within a unified spatial grid cell and continuous time window framework, the grid energy values ​​corresponding to each spatial grid cell within each time window are extracted to generate a time-series grid energy set. Within each time window, the spatial grid cells are topologically organized according to spatial adjacency relationships, establishing spatial adjacency associations between adjacent spatial grid cells to generate a time window spatial energy distribution structure. Between adjacent time windows, a temporally continuous association is established for the grid energy values ​​of the same spatial grid cell, generating a time-recursive energy association structure. The time window spatial energy distribution structure and the time-recursive energy association structure are jointly integrated to generate a spatiotemporal energy association structure that includes spatial adjacency associations and temporally continuous associations. The spatiotemporal energy association structures corresponding to each time window are continuously arranged according to the time window order to form a spatiotemporal diffused energy field. The spatiotemporal diffusion energy field is arranged in a matrix to form a spatiotemporal diffusion energy distribution matrix.

[0023] In this embodiment, the generation of the steady-state active domain and the diffusion active domain specifically includes: Extract the grid energy values ​​corresponding to each time window in the spatiotemporal diffusion energy distribution matrix; Within each time window, the grid energy values ​​of all spatial grid cells are statistically analyzed, and the average energy value within that time window is calculated. Within the same time window, the squared difference between the grid energy value and the average energy value of all spatial grid cells is calculated, and the squared differences are summed to obtain the energy fluctuation value. The energy threshold for the corresponding time window is determined based on the average energy value and the energy fluctuation value. Within the time window, the grid energy value of each spatial grid cell is compared with the energy threshold. When the grid energy value is less than or equal to the energy threshold, the spatial grid cell is divided into a steady-state active domain. When the grid energy value is greater than the energy threshold, the spatial grid cell is divided into a diffusion active domain. Within the time window, connectivity is determined for spatially adjacent spatial grid cells of the same type, forming a set of stable active domain connected regions and a set of diffuse active domain connected regions, respectively. The sets of connected regions of steady-state activity domains and the sets of connected regions of diffusion activity domains corresponding to each time window are combined in chronological order to form the community structure domain partitioning results.

[0024] In this embodiment, the generation of the community structure stability determination result specifically includes: Extract the set of spatial grid cells corresponding to the diffusion active domain and the grid energy values ​​in the spatiotemporal diffusion energy field to generate a diffusion active domain grid energy dataset. Within the same time window, a spatially connected network structure is established using spatial grid cells in the diffusion activity domain grid energy dataset as nodes, generating a set of spatially connected nodes and a set of connected edges; For any pair of adjacent nodes in the set of spatially connected nodes, calculate the difference in their corresponding grid energy values ​​to generate a set of path energy costs. In the spatial connectivity network structure, with the path energy cost set as a constraint, starting from the spatial grid cells in the diffusion activity domain, the connected nodes with the minimum path energy cost are selected step by step to generate the energy reachable path set; The generation of the set of energy-reachable paths specifically includes: Within the same time window, extract the set of spatially connected nodes and the set of connected edges. Mark the path energy cost for each connected edge in the edge set to form a weighted spatial connected network structure. In the weighted spatial connected network structure, each spatial grid cell in the diffusion activity domain is sequentially used as a starting node to establish a corresponding initial path set, and the cumulative path energy value in the initial path set is initialized to the grid energy value corresponding to the starting node. For each initial path set, traverse adjacent connected nodes in the weighted spatial connected network structure according to the direction of the connected edges, and compare the grid energy value corresponding to the adjacent connected nodes with the path energy cost value corresponding to the connected edges. Accumulate the energy values ​​to generate an expanded path set, and update the cumulative energy values ​​of the paths in the expanded path set. Within the expanded path set, compare the cumulative energy values ​​of multiple paths corresponding to the same termination node, retain the path with the smallest cumulative energy value, and delete the rest, forming a minimum energy path set. Repeat the path expansion and minimum energy path filtering operations until all reachable connected nodes have completed path updates, generating a minimum energy path network covering the diffusion activity domain. Number all paths in the minimum energy path network, extract the spatial grid cell sequence and cumulative energy value corresponding to each path, and generate an energy-reachable path set. For each energy-reachable path in the set of energy-reachable paths, the sequence of spatial grid cells it contains and the cumulative energy value of the corresponding grid cells are statistically analyzed to generate path cumulative energy data; The individual spatiotemporal trajectory sequence is mapped to a set of spatial grid cells to generate a set of the individual's actual activity paths; The overlap statistics of the set of actual activity paths of individuals and the set of energy-accessible paths are statistically analyzed at the spatial grid unit level to generate path overlap ratio data; Within the same time window, the cumulative energy data of the path and the path overlap ratio data are jointly calculated to generate a joint path determination value; The generation of the path joint decision value specifically includes: Within the same time window, cumulative path energy data and path overlap ratio data are extracted. The cumulative path energy data is sorted from smallest to largest to generate a path energy ranking sequence. For each path in the ranking sequence, the corresponding path overlap ratio value is extracted to form a path energy-overlap ratio correspondence sequence. For each path in the path energy-overlap ratio correspondence sequence, the energy deviation is calculated, which is the difference between the cumulative energy value of that path and the average cumulative energy value of all paths within the same time window. For each path, the structural consistency degree is calculated, which is the difference between the path overlap ratio value of that path and the average overlap ratio of all paths within the same time window. For each path, the energy deviation and structural consistency degree are matched by sign. When the signs of the energy deviation and structural consistency degree are the same, it is marked as a structurally consistent path; when the signs are different, it is marked as a structurally conflicting path. The number of structurally consistent paths and the number of structurally conflicting paths are counted, and the ratio between the number of structurally consistent paths and the total number of paths is calculated to generate a joint path determination value. The path joint determination value is compared with the joint determination threshold to generate the community structure stability determination result.

[0025] In this embodiment, the generation of path overlap ratio data specifically includes: Extract the spatial grid cell sequence corresponding to each individual within each time window from the set of actual activity paths of individuals, and generate a set of individual path grid sequence; Extract the spatial grid cell sequence of each energy reachable path in the energy reachable path set within the corresponding time window, and generate a reachable path grid sequence set; Within the same time window, for each individual path in the individual path grid sequence set, a spatial grid cell matching and statistical analysis is performed with each energy-reachable path in the reachable path grid sequence set to generate data on the number of overlapping grid cells. The ratio of the number of overlapping grid cells along a path to the length of the corresponding individual path grid sequence is calculated to generate a single path overlap ratio value. Within the same time window, the overlap ratio of all single paths is statistically analyzed to generate a set of overlap ratios for the time window. The overlap ratios of each time window are arranged in chronological order to form path overlap ratio data.

[0026] In this embodiment, the generation of the spatiotemporal evolution structure results of community footprints specifically includes: Extract the community structure stability determination results and the set of connected regions of steady-state activity domains and the set of connected regions of diffusion activity domains to form the structure domain dataset corresponding to the time window; Within each time window, coordinate aggregation is performed on the spatial grid cells in the set of connected regions of the steady-state activity domain to generate spatial distribution data of the community activity core area. Between adjacent time windows, the centroid coordinates of the spatial grid cells in the set of connected regions of the diffusion activity domain are calculated, and the difference between the centroid coordinates of adjacent time windows is calculated to generate diffusion trend direction data. The results of the community structure stability assessment are arranged in chronological order to generate a time series of structure stability. By combining the spatial distribution data of the community activity core area, the diffusion trend direction data, and the time series of structural stability according to the time window, the spatiotemporal evolution structure results of the community footprint are generated.

[0027] Example 1: To verify the feasibility of the present invention in practice, it was applied to a concentrated area of ​​wild Père David's deer in the coastal wetland reserve of Yancheng, Jiangsu Province. The area is open and includes mudflats, low vegetation zones, and seasonal waterholes. Wild Père David's deer move back and forth between the core habitat area and the outer foraging area in different seasons. Past monitoring methods mainly relied on manual patrols and infrared camera recordings. The population changes were judged by counting the number of occurrences and the range of activities. However, it was impossible to distinguish between the stable structure of the population activities and the diffusion and migration state. It was also difficult to continuously depict the evolution process of the core area of ​​the community activities. Especially during the transition between the spring breeding season and the autumn migration season, the range of population activities expands. Traditional methods often only conclude that the activity area has increased, but cannot explain whether this change is a stable expansion or a structural diffusion.

[0028] In this embodiment, high-position cameras and ground image acquisition devices are deployed in the wetland area to continuously collect video and surface images of the Père David's deer activity area. The acquisition time covers multiple consecutive time windows, and the acquisition range covers the core habitat area, buffer zone, and outer foraging zone. First, the continuous video data is processed by target detection, instance segmentation, and individual weight recognition to form an individual spatiotemporal trajectory sequence containing individual identity, spatial location coordinates, and time information. This trajectory sequence completely records the spatial movement paths of different individuals within multiple time windows, so that the group behavior is no longer presented as an overall outline, but is expressed in the form of traceable individual paths.

[0029] Footprint segmentation and 3D morphological reconstruction were performed on the surface image data. Image preprocessing eliminated the effects of light differences and surface humidity. The planar coordinates and 3D morphological features of the footprints were extracted and time-stamped with the acquisition time to form a footprint spatiotemporal distribution dataset. Since the soil in this wetland area is relatively soft, the elk footprints are preserved for a long time. The footprint data can supplement the activity information that was not continuously captured in the video, so that a spatial correlation is formed between the group's activity trajectory and the surface behavioral traces.

[0030] After obtaining the individual spatiotemporal trajectory sequence and footprint spatiotemporal distribution dataset, an individual-footprint association matrix is ​​established. The spatial proximity and temporal synchronization relationships between individual trajectories and footprints are matched. Stable association sets are selected within a continuous time window to form community footprint association results. These results not only reflect individual behavioral paths but also the distribution of behavioral traces left by individuals in a specific spatial region, laying the foundation for subsequent structural analysis.

[0031] The monitoring area is discretized into a grid under a unified spatial coordinate system, and the data is organized according to continuous time windows. The community footprint association results are mapped to a spatiotemporal discrete grid structure. The footprint frequency and individual activity frequency of each grid unit are statistically analyzed to generate a community footprint probability density field. Through this process, the changing trend of densely populated areas of group activity in different time windows can be observed intuitively. However, this invention does not stop at the density level, but further constructs a spatiotemporal diffusion energy field based on the change of probability density.

[0032] In the spatiotemporal diffusion energy field, grid energy values ​​are generated by combining grid probability density values, temporal variation intensity, and spatial diffusion intensity. All grid cells are spatiotemporally organized to form a spatiotemporal diffusion energy distribution matrix. During continuous observation in this wetland area, it was found that the grid energy values ​​of the outer foraging area showed a continuous upward trend in some time windows, while the grid energy values ​​of the core habitat area fluctuated less. By dividing the steady-state activity domain and diffusion activity domain by energy threshold, the stable habitat structure and diffusion migration structure can be clearly distinguished.

[0033] Within the diffusion activity domain, a spatial connectivity network structure is constructed. Based on the path energy cost, a set of energy-accessible paths is generated, and the spatiotemporal trajectory sequence of individuals is mapped to the grid path. The overlap ratio between the actual activity path of an individual and the energy-accessible path is calculated. At the same time, the path joint judgment value is generated by combining the cumulative energy of the path to obtain the community structure stability judgment result. Through this process, it is possible to determine whether the group activity continues to advance along the predetermined diffusion direction within a certain time window, or whether structural retreat or splitting occurs.

[0034] In practical applications, the stability of community structure is arranged through continuous time windows to form a time series of structural stability. Combined with the changes in the spatial centroid of the connected regions of the steady-state activity domain and the connected regions of the diffusion activity domain, the spatial distribution and diffusion trend of the community activity core area are generated. The observation results show that during the spring when foraging resources are abundant, the diffusion activity domain gradually advances to the outer mudflat area. However, when climate change or human disturbance occurs, the diffusion activity domain quickly shrinks back to the core habitat area. Under the method of this invention, this structural evolution process is presented as a continuous change in energy structure and path structure, rather than a single area change.

[0035] Through implementation, this invention effectively solves the problem that traditional methods cannot quantitatively determine the stability of community structure. It realizes a complete chain from individual behavior identification to community structure evolution analysis. This method maintains data consistency and structural coherence during long-term continuous monitoring, providing clear structural basis for judging the core area of ​​community activity and the direction of diffusion trends. It provides quantifiable decision-making reference for wetland protection and management. At the same time, due to the joint determination mechanism of energy structure and path structure, it can still stably identify the direction of community diffusion and stable structural areas under complex terrain and multi-path interference conditions, verifying the feasibility and practical value of this invention in real-world scenarios.

[0036] Table 1. Statistical Comparison of Community Footprint Spatiotemporal Structure Analysis Performance

[0037] As shown in Table 1, the method of this invention has achieved stable improvements in multiple structural determination-related indicators, while the improvement range is controlled within a reasonable range. The individual recognition accuracy has increased from 84.6% in the traditional trajectory statistics method to 88.7%, an increase of about 4 percentage points. This improvement mainly comes from the correlation processing between individual spatiotemporal trajectory and footprint spatiotemporal distribution data, which partially compensates for the impact of short-term occlusion or viewpoint changes on individual recognition. The footprint matching accuracy has increased from 72.8% and 76.5% to 82.4%, an increase of about 6 percentage points. In traditional methods, footprints are usually used as auxiliary information, while this invention constructs an individual-footprint correlation matrix to form a two-way constraint relationship between footprints and individual behavioral paths, reducing the false association caused by spatial proximity but temporal mismatch.

[0038] Regarding the positioning error in the core area of ​​the community, this invention reduces the error from 16.9 meters and 14.7 meters to 11.8 meters, a reduction of approximately 20% to 30%. This improvement does not rely solely on the peak spatial density, but rather combines the spatiotemporal diffusion energy field to divide the region, making the core area determination smoother and more stable, avoiding the impact of local high-frequency fluctuations on the overall judgment. The consistency rate of diffusion direction determination is increased from 58.3% and 63.9% to 74.6%, an improvement of more than 10 percentage points. Traditional methods usually rely on the overall trend of the trajectory for judgment, which is easily affected by individual discrete behaviors. In contrast, this invention constructs a structural reference path through a set of energy-accessible paths and corrects it by combining the path overlap ratio, making the diffusion direction determination more stable.

[0039] The misjudgment rate of structural stability decreased from 19.8% and 16.4% to 10.9%, a reduction of approximately 5 to 9 percentage points. This change stems from the joint calculation mechanism of path cumulative energy data and path overlap ratio data, which simultaneously constrains group activities at both the energy structure and behavioral path levels, reducing the judgment bias caused by misleading single indicators. The structural consistency of continuous time windows increased from 0.68 and 0.72 to 0.83, reflecting the enhanced stability of the invention in the time dimension. By constructing a spatiotemporal diffusion energy distribution matrix and organizing structural domain connectivity under multiple time windows, the structural evolution results no longer exhibit abrupt changes but rather form a continuous change process.

[0040] In a multipath interference environment, this invention reduces the judgment deviation rate from 22.5% and 18.7% to 12.6%, indicating that the structure judgment can still maintain high stability even when individual activity paths intersect or there is short-term abnormal movement. Although the processing time of a single time window increases from 4.5 seconds and 5.1 seconds to 5.9 seconds, the increase is limited and within an acceptable range. Overall, this invention achieves simultaneous improvement in the accuracy, stability, and anti-interference ability of community structure judgment while ensuring feasibility, verifying its effectiveness and application value in the spatiotemporal analysis of actual wild Père David's deer communities.

[0041] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying wild elk and spatiotemporal analyzing their footprints based on AI vision, characterized in that, Includes the following steps: Collect continuous video data and surface image data within the monitoring area, perform target detection, instance segmentation and individual weight recognition on the continuous video data, extract individual identity information, spatial location coordinates and timestamp information, and generate individual spatiotemporal trajectory sequences; Footprint segmentation and 3D morphology reconstruction are performed on surface image data to extract spatial coordinate information of footprints and generate a spatiotemporal distribution dataset of footprints. An individual-footprint association matrix is ​​established based on individual spatiotemporal trajectory sequences and footprint spatiotemporal distribution datasets to generate community footprint association results; A spatiotemporal discrete grid structure is constructed, and the community footprint association results are mapped to the spatiotemporal discrete grid structure. The footprint frequency and individual activity frequency of each grid cell are counted to generate a community footprint probability density field. The energy values ​​of each grid cell are calculated based on the probability density field of the community footprint, and a spatiotemporal diffusion energy field is constructed and a spatiotemporal diffusion energy distribution matrix is ​​generated. Based on the energy threshold, the spatiotemporal diffusion energy distribution matrix is ​​divided into regions to generate steady-state active domains and diffusion active domains; Under the constraints of steady-state and diffusion activity domains, a set of energy-accessible paths is generated. The degree of structural deviation of individual spatiotemporal trajectory sequences in the set of energy-accessible paths is calculated, and the community structure stability determination result is generated. Based on the community structure stability assessment results, the spatiotemporal evolution structure of the community footprint is generated, including the spatial distribution of the community activity core area, the direction of diffusion trend, and the temporal evolution of structural stability.

2. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 1, characterized in that, The generation of the footprint spatiotemporal distribution dataset specifically includes: The surface image data is subjected to image denoising, distortion correction and illumination equalization to generate preprocessed surface image data. Footprint region segmentation is performed on the preprocessed surface image data to generate footprint segmentation results; The footprint segmentation results are divided into connected components to generate a set of footprint regions; Calculate the planar coordinates of each footprint region in the footprint region set to generate footprint planar coordinate data; Three-dimensional morphological reconstruction of the footprint region set is performed based on surface image data to generate three-dimensional footprint morphological data. Morphological features are extracted from the three-dimensional morphological data of footprints to generate footprint morphological feature data; The footprint planar coordinate data, footprint three-dimensional morphological data, and footprint morphological feature data are time-stamped and uniformly organized according to the collection time to generate a footprint spatiotemporal distribution dataset.

3. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 1, characterized in that, The generation of the community footprint association results specifically includes: The individual spatiotemporal trajectory sequences are sorted by time and standardized by spatial coordinates to form a set of individual trajectories containing individual identity information, spatial location coordinates and corresponding timestamp information; The footprint spatiotemporal distribution dataset is processed by coordinate unification and time stamping to form a footprint set containing footprint spatial coordinates and corresponding collection times; For each individual identifier in the individual trajectory set, the spatial coordinates of the footprints in the spatially adjacent area are retrieved within the corresponding timestamp range. The spatial proximity relationship between the individual's spatial location and the footprint's spatial location is calculated, and the time synchronization relationship between the individual's timestamp and the footprint collection time is calculated. When the spatial proximity relationship meets the preset spatial association condition and the temporal synchronization relationship meets the preset temporal association condition, it is determined that there is an association between the individual's identity and the corresponding footprint. All individuals and footprints that meet the association conditions are labeled, and an individual-footprint association matrix is ​​constructed. The rows of the individual-footprint association matrix correspond to the individual identity identifier, the columns correspond to the footprint number, and the matrix elements represent the association status between the corresponding individual and the footprint. The individual-footprint association matrix is ​​cumulatively statistically analyzed within a continuous time window, and individual-footprint pairs that continuously meet the association conditions within multiple continuous time windows are selected to form a stable association set. Community footprint association results are generated based on stable association sets, including the footprint sets corresponding to each individual's identity and their temporal distribution information.

4. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 1, characterized in that, The generation of the community footprint probability density field specifically includes: Under a unified spatial coordinate system, the monitoring area is discretized into grids, and multiple spatial grid units are divided according to a preset spatial resolution. Each spatial grid unit corresponds to a unique spatial range and has a definite grid number. The timeline is divided according to a preset time window length to form a continuous time window sequence; Within each time window, the community footprint association results are mapped to spatial grid cells, the number of footprints in each spatial grid cell is counted, and footprint frequency data is generated. Within the same time window, the number of individual activities in each spatial grid unit is counted based on the community footprint association results, and individual activity frequency data is generated. For each spatial grid cell, within the corresponding time window, the footprint frequency data and individual activity frequency data are weighted according to preset weights to generate a joint frequency value; The joint frequency values ​​of all spatial grid cells within the same time window are normalized to generate the grid probability density values ​​for the corresponding time window. The grid probability density values ​​of each time window are combined in chronological order to generate a community footprint probability density field.

5. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 1, characterized in that, The generation of the spatiotemporal diffusion energy distribution matrix specifically includes: Under the framework of unified spatial grid cells and continuous time windows, the probability density field of community footprints is extracted to obtain the grid probability density value of each spatial grid cell within each time window. For each spatial grid cell, the change in grid probability density value between adjacent time windows is calculated to generate time variation intensity data; For each spatial grid cell, the grid probability density difference between it and its adjacent spatial grid cells is calculated within the same time window to generate spatial diffusion intensity data; In each spatial grid cell and time window, the grid probability density value, time variation intensity data, and spatial diffusion intensity data are combined according to preset weights to generate the corresponding grid energy value. The grid energy values ​​are organized according to the spatial grid cells and time windows to construct a spatiotemporal diffusion energy field; The spatiotemporal diffusion energy field is arranged in a matrix to form a spatiotemporal diffusion energy distribution matrix.

6. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 1, characterized in that, The generation of the steady-state active domain and the diffusion active domain specifically includes: Extract the grid energy values ​​corresponding to each time window in the spatiotemporal diffusion energy distribution matrix; Within each time window, the grid energy values ​​of all spatial grid cells are statistically analyzed, and the average energy value within that time window is calculated. Within the same time window, the squared difference between the grid energy value and the average energy value of all spatial grid cells is calculated, and the squared differences are summed to obtain the energy fluctuation value. The energy threshold for the corresponding time window is determined based on the average energy value and the energy fluctuation value. Within the time window, the grid energy value of each spatial grid cell is compared with the energy threshold. When the grid energy value is less than or equal to the energy threshold, the spatial grid cell is divided into a steady-state active domain. When the grid energy value is greater than the energy threshold, the spatial grid cell is divided into a diffusion active domain. Within the time window, connectivity is determined for spatially adjacent spatial grid cells of the same type, forming a set of stable active domain connected regions and a set of diffuse active domain connected regions, respectively. The sets of connected regions of steady-state activity domains and the sets of connected regions of diffusion activity domains corresponding to each time window are combined in chronological order to form the community structure domain partitioning results.

7. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 1, characterized in that, The generation of the community structure stability determination result specifically includes: Extract the set of spatial grid cells corresponding to the diffusion active domain and the grid energy values ​​in the spatiotemporal diffusion energy field to generate a diffusion active domain grid energy dataset. Within the same time window, a spatially connected network structure is established using spatial grid cells in the diffusion activity domain grid energy dataset as nodes, generating a set of spatially connected nodes and a set of connected edges; For any pair of adjacent nodes in the set of spatially connected nodes, calculate the difference in their corresponding grid energy values ​​to generate a set of path energy costs. In the spatial connectivity network structure, with the path energy cost set as a constraint, starting from the spatial grid cells in the diffusion activity domain, the connected nodes with the minimum path energy cost are selected step by step to generate the energy reachable path set; For each energy-reachable path in the set of energy-reachable paths, the sequence of spatial grid cells it contains and the cumulative energy value of the corresponding grid cells are statistically analyzed to generate path cumulative energy data; The individual spatiotemporal trajectory sequence is mapped to a set of spatial grid cells to generate a set of the individual's actual activity paths; The overlap statistics of the set of actual activity paths of individuals and the set of energy-accessible paths are statistically analyzed at the spatial grid unit level to generate path overlap ratio data; Within the same time window, the cumulative energy data of the path and the path overlap ratio data are jointly calculated to generate a joint path determination value; The path joint determination value is compared with the joint determination threshold to generate the community structure stability determination result.

8. The method for identifying wild elk and spatiotemporal analysis of herd footprints based on AI vision according to claim 7, characterized in that, The generation of the path overlap ratio data specifically includes: Extract the spatial grid cell sequence corresponding to each individual within each time window from the set of actual activity paths of individuals, and generate a set of individual path grid sequence; Extract the spatial grid cell sequence of each energy reachable path in the energy reachable path set within the corresponding time window, and generate a reachable path grid sequence set; Within the same time window, for each individual path in the individual path grid sequence set, a spatial grid cell matching and statistical analysis is performed with each energy-reachable path in the reachable path grid sequence set to generate data on the number of overlapping grid cells. The ratio of the number of overlapping grid cells along a path to the length of the corresponding individual path grid sequence is calculated to generate a single path overlap ratio value. Within the same time window, the overlap ratio of all single paths is statistically analyzed to generate a set of overlap ratios for the time window. The overlap ratios of each time window are arranged in chronological order to form path overlap ratio data.

9. A method for identifying wild elk and spatiotemporal analyzing their footprints based on AI vision, as described in claim 1, is characterized in that... The generation of the spatiotemporal evolution structure results of the community footprint specifically includes: Extract the community structure stability determination results and the set of connected regions of steady-state activity domains and the set of connected regions of diffusion activity domains to form the structure domain dataset corresponding to the time window; Within each time window, coordinate aggregation is performed on the spatial grid cells in the set of connected regions of the steady-state activity domain to generate spatial distribution data of the community activity core area. Between adjacent time windows, the centroid coordinates of the spatial grid cells in the set of connected regions of the diffusion activity domain are calculated, and the difference between the centroid coordinates of adjacent time windows is calculated to generate diffusion trend direction data. The results of the community structure stability assessment are arranged in chronological order to generate a time series of structure stability. By combining the spatial distribution data of the community activity core area, the diffusion trend direction data, and the time series of structural stability according to the time window, the spatiotemporal evolution structure results of the community footprint are generated.

Images