Bird flight path intelligent tracking and monitoring method and system based on big data analysis

Abstract

The application discloses a bird flight path intelligent tracking and monitoring method and system based on big data analysis, S1. generating an original bird observation data set; S2. output standardized bird observation data set; S3. based on the standardized bird observation data set, a multi-target tracking algorithm is called to obtain a bird trajectory segment data set; S4. a lion group optimization search model is constructed, and the bird trajectory segment data set is input into the lion group optimization search model to form a candidate trajectory data set; S5. the candidate trajectory data set is input into a space-time Transformer model configured with a rotating position embedding to generate a continuous flight path data set; S6. output a cross-region bird migration monitoring result set. The application can cover bird migration continuous tracking requirements under different time periods, multiple geographical zones and heterogeneous device collection conditions, effectively support the application scenarios of ecological protection, migration research and airspace control, and have remarkable engineering promotion and ecological protection value.

Description

Technical Field

[0001] This invention relates to the field of bird monitoring technology, and in particular to a method and system for intelligent tracking and monitoring bird flight paths based on big data analysis. Background Technology

[0002] With the continuous advancement of ecological monitoring technology, cross-regional continuous tracking of bird migration routes has become an important supporting means in the fields of global biodiversity conservation, climate change research, and aviation safety management. Currently, bird migration route monitoring mainly relies on a variety of observation technologies, such as satellite transmitters, ground radar stations, video monitoring equipment, and acoustic sensors, deployed at key nodes. By collecting and stitching together the bird's activity trajectories, the migration dynamics can be recorded.

[0003] With the application of big data and artificial intelligence technologies, researchers have attempted to use recurrent neural networks and self-attention Transformer sequence modeling algorithms to align bird trajectories and interpolate missing segments. However, existing models generally use absolute position encoding, which is difficult to handle migration trajectories with long spans, sparse nodes, and drastic changes in geographical location. This can easily lead to a decline in long-distance dependent capture capabilities, large interpolation errors for missing segments, and unstable identity preservation. At the same time, existing data fusion methods often ignore the ecological constraints in bird migration processes, such as wind direction, topography, physical and biological boundaries of species differences, and cannot achieve intelligent optimization and reliable evaluation of the trajectory splicing process. Furthermore, existing technologies have not yet achieved the linkage and normalization processing of long-sequence, multi-target, cross-regional, and heterogeneous observation data, making it difficult to output high-confidence, traceable, and continuous migration path results.

[0004] In conclusion, there is an urgent need to develop a new method for continuous cross-regional tracking of bird flight paths that can effectively solve the above problems. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent tracking and monitoring method and system for bird flight paths based on big data analysis. This invention can cover the continuous tracking needs of bird migration under different time periods, multiple geographical zones, and heterogeneous equipment collection conditions, effectively supporting application scenarios such as ecological protection, migration research, and airspace management, and has significant engineering promotion and ecological protection value.

[0006] A method for intelligent tracking and monitoring of bird flight paths based on big data analysis according to an embodiment of the present invention includes the following steps:

[0007] S1. Deploy bird monitoring equipment in multiple geographical areas such as plateaus, mountains, wetlands and coasts to collect multimodal bird observation data including images, videos and sounds in real time and generate raw bird observation datasets;

[0008] S2. Perform data preprocessing on the original bird observation dataset, use target detection and recognition algorithms to locate and identify bird targets in images and videos, use audio recognition models to identify bird species corresponding to sound signals, unify time and space coordinates based on GPS labels, remove low-confidence or duplicate observation records and clean outliers, and output a standardized bird observation dataset.

[0009] S3. Based on the standardized bird observation dataset, a multi-target tracking algorithm is called to establish the temporal trajectory of the same bird in consecutive video frames. The image, video and sound recognition results are fused to construct a set of trajectory segments, and a bird trajectory segment dataset is obtained.

[0010] S4. Construct a lion pride optimization search model. Input the bird trajectory fragment dataset into the lion pride optimization search model. Complete the multi-objective coarse concatenation through global search by male lions, local screening by female lions, and cub diffusion exploration. Output the candidate trajectory set and confidence matrix to form a candidate trajectory dataset.

[0011] S5. Input the candidate trajectory dataset into the spatiotemporal Transformer model configured with rotational position embedding, perform long-span sequence alignment, bird individual identification and missing segment interpolation based on rotational geographic angular distance encoding, and obtain a continuous and complete set of flight paths to generate a continuous flight path dataset;

[0012] S6. After evaluating the error of the continuous flight path dataset and updating the weight parameters based on the evaluation results, write the data into the trajectory database to form a stored continuous flight path dataset. Then, associate the stored continuous flight path dataset with the bird ecological knowledge graph and real-time environmental data to output a set of cross-regional bird migration monitoring results.

[0013] Optionally, S2 includes the following steps:

[0014] S21. Perform bird target detection and recognition processing on the image and video data in the original bird observation dataset, and output the set of spatial coordinates of all detected bird targets in each frame of image or video. Each spatial coordinate in the set consists of an abscissa, a ordinate, and a corresponding time label. The number of spatial coordinate sets is equal to the number of bird targets detected in the image or video frame.

[0015] S22. Use an audio recognition model to identify bird species from the sound data in the original bird observation dataset. Extract the voiceprint features of each sound data segment and match the voiceprint features with known bird species labels to generate an audio recognition result set. Each element in the audio recognition result set includes the bird species label and time label corresponding to the sound data. The number of audio recognition result sets is equal to the number of sound data segments.

[0016] S23. Based on the GPS labels in the original bird observation dataset, all observation data are processed by time normalization and spatial coordinate normalization. Each observation data is unified into standardized observation data after passing through the spatial coordinate normalization function and the time normalization function. The standardized observation data includes the normalized x-coordinate, y-coordinate, normalized time label and observation data category. The number of standardized observation data sets is equal to the total number of observation data.

[0017] S24. Perform confidence screening and duplicate observation removal on the standardized observation dataset. Set a confidence threshold, retain only the observation data with a confidence score not lower than the confidence threshold, and remove duplicate observation records with the same time label and spatial coordinates. Each element in the confidence screening set is a unique observation data with a confidence score not lower than the confidence threshold.

[0018] S25. Perform outlier cleaning on the confidence screening set, calculate the Euclidean distance between each observation data and the mean vector in the confidence screening set, retain only the observation data whose Euclidean distance is not higher than the outlier discrimination threshold, and finally output the standardized bird observation dataset.

[0019] Optionally, S3 includes the following steps:

[0020] S31. For the standardized bird observation dataset, a multi-target tracking algorithm is used to assign the identity of each bird target in continuous video frames. In each frame, each spatial coordinate in the spatial coordinate set is processed according to its horizontal coordinate, vertical coordinate and time label. By comparing the correlation between the spatial coordinates and time labels of adjacent frames, the spatial coordinates of the same bird in different frames are associated to form an initial time-series trajectory set. Each time-series trajectory in the initial time-series trajectory set consists of several spatial coordinates and time labels that have been time-series associated. The number of time-series trajectories in the initial time-series trajectory set is equal to the total number of trajectories.

[0021] S32. For the audio recognition result set in the standardized bird observation dataset, extract the bird species label and time label corresponding to each sound data segment, match and associate the bird species label identified under the same time label with the spatial coordinates matched in the video frame, fuse the image, video and sound recognition results, and construct a fused observation segment set containing spatial coordinates, time labels and bird species labels. Each fused observation segment in the fused observation segment set consists of spatial coordinates, time labels and bird species labels, and the number of fused observation segment sets is equal to the number of fused segments.

[0022] S33. For the initial time-series trajectory set and the fused observation segment set, trajectory fusion is performed based on the trajectory similarity index function. The trajectory similarity index function is used to measure the comprehensive similarity between the initial time-series trajectory and the fused observation segment in terms of spatial coordinates and species labels. Each trajectory segment in the trajectory segment set consists of time-related spatial coordinates, time labels, and bird species labels. The number of trajectory segment sets is equal to the number of trajectory segments.

[0023] S34. Perform uniform formatting on all trajectory segments in the trajectory segment set so that each trajectory segment contains temporally associated spatial coordinates, time labels, and bird species labels, and output the bird trajectory segment dataset.

[0024] Optionally, the trajectory similarity index function value is obtained by calculating the Euclidean distance between the initial time-series trajectory and the fused observation segment in spatial coordinates, and then calculating the consistency measure between the initial time-series trajectory and the fused observation segment in species label. The Euclidean distance and the consistency measure are multiplied by the corresponding normalized weight coefficients and then accumulated. The accumulated value is obtained by taking the reciprocal and adding one. By maximizing the trajectory similarity index function value, the trajectories and observation segments of the same bird are fused to obtain a set of trajectory segments.

[0025] Optionally, S4 includes the following steps:

[0026] S41. Construct a lion pride optimization search model based on cross-regional migration constraints, taking the bird trajectory fragment dataset as input. The bird trajectory fragment dataset is divided into geographical zones, and each geographical zone corresponds to a unique number.

[0027] S42. For any two bird trajectory segments, calculate the spatial distance between the two bird trajectory segments, the time label difference between the two bird trajectory segments, and the species label consistency parameter between the two bird trajectory segments by assigning spatial distance weight, time difference weight, and species consistency weight. Then, sum the above three values ​​according to the spatial distance weight, time difference weight, and species consistency weight to obtain the basic fitness parameter. The basic fitness parameter is used to measure the ecological splicing cost between any two bird trajectory segments. The spatial distance is measured by calculating the geographical distance between the end point of the first bird trajectory segment and the start point of the second bird trajectory segment. The time label difference is the absolute value of the end time label of the first bird trajectory segment and the start time label of the second bird trajectory segment. The species consistency parameter is zero when the species labels of the two bird trajectory segments are the same and one when they are different. The sum of the spatial distance weight, time difference weight, and species consistency weight is one.

[0028] S43. Based on the basic fitness parameters, wind direction matching parameters, terrain elevation constraint parameters, and historical migration record consistency parameters are introduced to reflect the matching of the angle between wind direction and flight direction between two bird trajectory segments, the terrain elevation jump between two bird trajectory segments, and the splicing consistency of two bird trajectory segments in historical data, respectively. By assigning wind direction matching correction weights, terrain elevation correction weights, and historical migration record correction weights, the wind direction matching parameters, terrain elevation constraint parameters, and historical migration record consistency parameters are added as correction terms to the basic fitness parameters to obtain enhanced fitness parameters. Enhanced fitness parameters are used to measure the comprehensive splicing cost of two bird trajectory segments under ecological and physical constraints. The closer the value of the wind direction matching parameter is to one, the more consistent the wind direction and flight direction are. The terrain elevation constraint parameter is used to determine whether the terrain elevation difference exceeds the range that the species can fly over. The historical migration record consistency parameter is used to determine whether the splicing instance exists in the historical data.

[0029] S44. The male lion performs a global search in the bird trajectory fragment dataset. For each geographic zone, all bird trajectory fragment pairs in the geographic zone are combined and matched. The enhancement fitness parameter of each bird trajectory fragment pair is calculated. The bird trajectory fragment pairs with the lowest enhancement fitness parameters are taken as the preliminary candidate pairing results. The preliminary candidate pairing results form the global candidate pairing set.

[0030] S45. Based on the global candidate pairing set, a migration ecological filtering mechanism is introduced by the female lion. For each candidate pairing, it is determined in turn whether its time tag interval is less than or equal to the maximum allowable migration time jump threshold, whether the wind direction matching parameter is greater than or equal to the lower limit of the wind-assisted effect, and whether the terrain climbing constraint parameter is equal to zero. Only when all conditions are met is the pairing retained as a locally effective pairing. All locally effective pairs form a local candidate pairing set.

[0031] S46. For unmatched bird trajectory segments in the bird trajectory segment dataset, introduce temporal and spatial perturbations to randomly perturb the time labels and spatial coordinates of the cubs to generate new combinations, and calculate the enhanced fitness parameter of each combination. Apply the same screening criteria as above to all new combinations, retain valid pairs, and form a perturbation expansion candidate set for all valid pairs.

[0032] S47. All valid pairs in the local candidate pairing set and the perturbation expansion candidate set are spliced ​​together in chronological order according to the time label to obtain a complete candidate trajectory. Each complete candidate trajectory consists of several sequentially spliced ​​bird trajectory segments. During the trajectory splicing process, the enhancement fitness parameters between each group of bird trajectory segments are recorded. The arithmetic mean of all enhancement fitness parameters is used as the average path splicing cost of the complete candidate trajectory.

[0033] S48. For each complete candidate trajectory, based on the path splicing average cost, the path splicing average cost is converted into a confidence parameter using a negative normalized exponential function. The standardization factor is used to adjust the distribution range of the confidence parameter. The confidence parameters of all complete candidate trajectories form a scoring matrix. All complete candidate trajectories and the scoring matrix are output to form a candidate trajectory dataset.

[0034] Optionally, S5 includes the following steps:

[0035] S51. The candidate trajectory dataset consists of several candidate trajectories. Each candidate trajectory is composed of several sequentially spliced ​​bird trajectory segments. Each bird trajectory segment contains spatial coordinates, time labels, and bird species labels. The candidate trajectory dataset is transformed into a sequence structure that the model can recognize, resulting in a set of candidate trajectory sequences. Each candidate trajectory sequence is a time-ordered sequence.

[0036] S52. For the candidate trajectory sequence set, calculate the spatial variation based on the longitude and latitude of each trajectory point, and convert the spatial variation into polar coordinates to obtain the geographic argument parameter and geographic modulus parameter. The geographic argument parameter is used to measure the spatial angular distance between two trajectory points, and the geographic modulus parameter is used to measure the geographic distance between two trajectory points. The geographic argument parameter and geographic modulus parameter, together with the time label difference, form a rotated geographic angular distance code. The rotated geographic angular distance code is used as the spatial and temporal feature embedding of the candidate trajectory sequence, and the rotated geographic angular distance code corresponds one-to-one with each trajectory point.

[0037] S53. The candidate trajectory sequence and its corresponding rotation geographic distance code are input into the spatiotemporal Transformer model configured with rotation position embedding. The spatiotemporal Transformer model directly uses the geographic argument parameter and geographic magnitude parameter of the rotation geographic distance code in each attention head to measure the spatial relationship between trajectory points, and combines the time label difference as a measure of temporal relationship. In the process of calculating attention weight, the attention distribution ratio between different trajectory points is dynamically adjusted based on the geographic argument parameter and geographic magnitude parameter of the rotation geographic distance code, so as to realize multi-dimensional collaborative perception of space and time, and output a multi-head attention weight matrix. All multi-head attention weight matrices correspond one-to-one with the candidate trajectory sequence.

[0038] S54. Based on the multi-head attention weight matrix, perform long-span sequence alignment for each candidate trajectory sequence. During the sequence alignment process, for each trajectory segment and its context segment, use the geographic argument parameter and geographic modulus parameter of the rotation geographic distance encoding to dynamically adjust the temporal arrangement order of trajectory points and optimize the spatial connection structure between trajectory points to ensure that the aligned trajectory sequence maintains continuity and rationality in both spatial and temporal dimensions, and generate an aligned trajectory sequence set.

[0039] S55. Based on the aligned trajectory sequence set, bird individual identification is performed based on the geographic argument parameter, geographic modulus parameter, and time tag difference of the rotation geographic distance encoding of each trajectory point. By comparing the geographic argument parameter, geographic modulus parameter, and time tag difference between different trajectory points in the same trajectory sequence, the spatial proximity, temporal continuity, and motion consistency of the trajectory points are comprehensively judged. Trajectory points belonging to the same bird are aggregated into a unified identity, resulting in a set of labeled trajectory identity tags. The set of trajectory identity tags corresponds one-to-one with the aligned trajectory sequence set.

[0040] S56. For the missing segment regions in the aligned trajectory sequence set, the geographic argument parameter and geographic modulus parameter of the rotated geographic angle distance encoding and their corresponding time tag difference are used. Based on the interpolation prediction method, the spatial coordinate prediction value and time tag prediction value of the missing segment are calculated. The prediction value is added to the missing segment region to realize the continuity repair of the trajectory sequence and generate a continuous and complete flight path set.

[0041] S57. All continuous and complete flight path sets are uniformly formatted. Each continuous and complete flight path contains temporally associated spatial coordinates, time labels, bird species labels, trajectory identity labels, and rotational geographic distance codes corresponding to each trajectory point. The continuous flight path dataset is then output.

[0042] Optionally, S6 includes the following steps:

[0043] S61. Write the continuous flight path dataset into the trajectory database. The trajectory database records the spatial coordinates, time label, bird species label, trajectory identity label, rotation geographic distance code and weight parameters of each flight path to form a stored continuous flight path dataset.

[0044] S62. Associate the stored continuous flight path dataset with the bird ecological knowledge graph. Based on the spatial coordinates, time labels and bird species labels of each flight path, extract the corresponding habitat type, migratory species, migration time window and ecological risk level parameters from the bird ecological knowledge graph to generate a path-knowledge graph matching parameter set.

[0045] S63. Associate the stored continuous flight path dataset with real-time environmental data. Based on the spatial coordinates, time labels and weight parameters of each flight path, extract the corresponding meteorological conditions, surface features and environmental disturbance parameters from the real-time environmental data to generate a path-environment parameter set.

[0046] S64. Based on the path-knowledge graph matching parameter set and the path-environment parameter set, output the cross-regional bird migration monitoring result set, and output the set of all cross-regional bird migration monitoring results.

[0047] Optionally, the cross-regional bird migration monitoring results set includes the following result types:

[0048] The habitat type identification results are determined based on the spatial coordinates of the flight path and the habitat type parameters in the knowledge graph, generating a habitat type label for each flight path;

[0049] Based on the identification results of high-risk migration areas, the spatiotemporal distribution monitoring results of species in the flight path and knowledge graph, and the bird species tags, time tags and spatial coordinates in the environmental data, the spatiotemporal distribution change trends of each species are output.

[0050] Based on the spatial overlap and temporal consistency parameters of multiple flight paths, the migration corridor identification results output migration corridor segments and corridor activity indicators.

[0051] A bird flight path intelligent tracking and monitoring system based on big data analysis, used to execute a bird flight path intelligent tracking and monitoring method based on big data analysis, includes the following modules:

[0052] The multimodal bird observation data acquisition module is used to acquire multimodal bird observation data, including images, videos, and sounds, in real time and generate raw bird observation datasets.

[0053] The data preprocessing module is used to perform data preprocessing on the original bird observation dataset, locate and identify bird targets in images and videos, identify bird species corresponding to sound signals, and output a standardized bird observation dataset.

[0054] The trajectory segment modeling module is used to establish the temporal trajectory of the same bird based on a standardized bird observation dataset, call a multi-target tracking algorithm, fuse multimodal recognition results, and construct a trajectory segment dataset.

[0055] The lion pride optimization module is used to perform lion pride optimization search on the trajectory segment dataset based on cross-regional migration constraints. It uses spatial distance weight, temporal difference weight, species consistency weight and ecological correction parameters to obtain candidate trajectory datasets and confidence score matrices.

[0056] The Spatiotemporal Transformer fine-grained alignment module is used to input candidate trajectory datasets into a spatiotemporal Transformer model configured with rotational geographic distance encoding. Based on the rotational geographic distance encoding, it performs long-span sequence alignment, bird individual identification, and missing segment interpolation to generate a continuous flight path dataset.

[0057] The trajectory evaluation and knowledge fusion module is used to evaluate the error of the continuous flight path dataset and update the weight parameters based on the evaluation results before writing them into the trajectory database. It associates the stored continuous flight path dataset with the bird ecological knowledge graph and real-time environmental data, and outputs a set of cross-regional bird migration monitoring results.

[0058] The beneficial effects of this invention are:

[0059] (1) In the lion pride optimization search model, this invention combines geographical zone zoning, wind direction matching, terrain climbing constraints and ecological characteristics of historical migration record consistency to carry out multi-level constraints and corrections on global pairing and local screening of trajectory fragments, realizing intelligent coarse concatenation of trajectory fragments across multiple targets and regions. Compared with traditional splicing methods based on heuristic thresholds or Euclidean distance, it can dynamically adapt to the migration behavior under different species, geographical environment and climate conditions, effectively reducing the probability of mis-sponging and missing splicing.

[0060] (2) This invention proposes a rotation geographic distance encoding method that combines spatial argument parameters and geographic modulus parameters with time tag difference. It embeds this method into a multi-head attention mechanism to achieve dynamic correlation in the spatial and temporal dimensions of the trajectory sequence. This method can accurately model bird flight paths under long-distance and complex geographic conditions. The Transformer can capture long-distance dependence and periodic migration patterns in sequence alignment, individual identity identification and missing segment interpolation, achieving a reconstruction result with controllable error and coherent trajectory.

[0061] (3) This invention combines multi-source observation data (images, sounds, trajectories, and meteorological data) with a dynamic weight update mechanism, and integrates knowledge graph association with environmental data fusion. It can output real-time results of bird habitat type identification, high-risk migration area identification, species spatiotemporal distribution monitoring, and multi-dimensional monitoring of migration corridor activity. In actual deployment, it exhibits extremely high adaptability and scalability, and can cover the continuous tracking needs of bird migration under different time periods, multiple geographical zones, and heterogeneous equipment collection conditions. It effectively supports application scenarios of ecological protection, migration research, and airspace management, and has significant engineering promotion and ecological protection value. Attached Figure Description

[0062] 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:

[0063] Figure 1 This is a flowchart of a bird flight path intelligent tracking and monitoring method and system based on big data analysis proposed in this invention. Detailed Implementation

[0064] 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.

[0065] refer to Figure 1 A method for intelligent tracking and monitoring of bird flight paths based on big data analysis includes the following steps:

[0066] S1. Deploy bird monitoring equipment in multiple geographical areas such as plateaus, mountains, wetlands and coasts to collect multimodal bird observation data including images, videos and sounds in real time and generate raw bird observation datasets;

[0067] S2. Perform data preprocessing on the original bird observation dataset, use target detection and recognition algorithms to locate and identify bird targets in images and videos, use audio recognition models to identify bird species corresponding to sound signals, unify time and space coordinates based on GPS labels, remove low-confidence or duplicate observation records and clean outliers, and output a standardized bird observation dataset.

[0068] S3. Based on the standardized bird observation dataset, a multi-target tracking algorithm is called to establish the temporal trajectory of the same bird in consecutive video frames. The image, video and sound recognition results are fused to construct a set of trajectory segments, and a bird trajectory segment dataset is obtained.

[0069] S4. Construct a lion pride optimization search model. Input the bird trajectory fragment dataset into the lion pride optimization search model. Complete the multi-objective coarse concatenation through global search by male lions, local screening by female lions, and cub diffusion exploration. Output the candidate trajectory set and confidence matrix to form a candidate trajectory dataset.

[0070] S5. Input the candidate trajectory dataset into the spatiotemporal Transformer model configured with rotational position embedding, perform long-span sequence alignment, bird individual identification and missing segment interpolation based on rotational geographic angular distance encoding, and obtain a continuous and complete set of flight paths to generate a continuous flight path dataset;

[0071] S6. After evaluating the error of the continuous flight path dataset and updating the weight parameters based on the evaluation results, write the data into the trajectory database to form a stored continuous flight path dataset. Then, associate the stored continuous flight path dataset with the bird ecological knowledge graph and real-time environmental data to output a set of cross-regional bird migration monitoring results.

[0072] In this embodiment, S2 includes the following steps:

[0073] S21. Perform bird target detection and recognition processing on the image and video data in the original bird observation dataset, and output the set of spatial coordinates of all detected bird targets in each frame of image or video. Each spatial coordinate in the set consists of an abscissa, a ordinate, and a corresponding time label. The number of spatial coordinate sets is equal to the number of bird targets detected in the image or video frame.

[0074] S22. Use an audio recognition model to identify bird species from the sound data in the original bird observation dataset. Extract the voiceprint features of each sound data segment and match the voiceprint features with known bird species labels to generate an audio recognition result set. Each element in the audio recognition result set includes the bird species label and time label corresponding to the sound data. The number of audio recognition result sets is equal to the number of sound data segments.

[0075] S23. Based on the GPS labels in the original bird observation dataset, all observation data are processed by time normalization and spatial coordinate normalization. Each observation data is unified into standardized observation data after passing through the spatial coordinate normalization function and the time normalization function. The standardized observation data includes the normalized x-coordinate, y-coordinate, normalized time label and observation data category. The number of standardized observation data sets is equal to the total number of observation data.

[0076] S24. Perform confidence screening and duplicate observation removal on the standardized observation dataset. Set a confidence threshold, retain only the observation data with a confidence score not lower than the confidence threshold, and remove duplicate observation records with the same time label and spatial coordinates. Each element in the confidence screening set is a unique observation data with a confidence score not lower than the confidence threshold.

[0077] S25. Perform outlier cleaning on the confidence screening set, calculate the Euclidean distance between each observation data and the mean vector in the confidence screening set, retain only the observation data whose Euclidean distance is not higher than the outlier discrimination threshold, and finally output the standardized bird observation dataset.

[0078] In this embodiment, S3 includes the following steps:

[0079] S31. For the standardized bird observation dataset, a multi-target tracking algorithm is used to assign the identity of each bird target in continuous video frames. In each frame, each spatial coordinate in the spatial coordinate set is processed according to its horizontal coordinate, vertical coordinate and time label. By comparing the correlation between the spatial coordinates and time labels of adjacent frames, the spatial coordinates of the same bird in different frames are associated to form an initial time-series trajectory set. Each time-series trajectory in the initial time-series trajectory set consists of several spatial coordinates and time labels that have been time-series associated. The number of time-series trajectories in the initial time-series trajectory set is equal to the total number of trajectories.

[0080] S32. For the audio recognition result set in the standardized bird observation dataset, extract the bird species label and time label corresponding to each sound data segment, match and associate the bird species label identified under the same time label with the spatial coordinates matched in the video frame, fuse the image, video and sound recognition results, and construct a fused observation segment set containing spatial coordinates, time labels and bird species labels. Each fused observation segment in the fused observation segment set consists of spatial coordinates, time labels and bird species labels, and the number of fused observation segment sets is equal to the number of fused segments.

[0081] S33. For the initial time-series trajectory set and the fused observation segment set, trajectory fusion is performed based on the trajectory similarity index function. The trajectory similarity index function is used to measure the comprehensive similarity between the initial time-series trajectory and the fused observation segment in terms of spatial coordinates and species labels. Each trajectory segment in the trajectory segment set consists of time-related spatial coordinates, time labels, and bird species labels. The number of trajectory segment sets is equal to the number of trajectory segments.

[0082] S34. Perform uniform formatting on all trajectory segments in the trajectory segment set so that each trajectory segment contains temporally associated spatial coordinates, time labels, and bird species labels, and output the bird trajectory segment dataset.

[0083] In this embodiment, the trajectory similarity index function value is obtained by calculating the Euclidean distance between the initial time-series trajectory and the fused observation segment in spatial coordinates, and then calculating the consistency measure between the initial time-series trajectory and the fused observation segment in species label. The Euclidean distance and the consistency measure are multiplied by the corresponding normalized weight coefficients and then summed. The sum is obtained by taking the reciprocal of the sum and adding one. By maximizing the trajectory similarity index function value, the trajectories and observation segments of the same bird are fused to obtain a set of trajectory segments.

[0084] In this embodiment, S4 includes the following steps:

[0085] S41. Construct a lion pride optimization search model based on cross-regional migration constraints, taking the bird trajectory fragment dataset as input. The bird trajectory fragment dataset is divided into geographical zones, and each geographical zone corresponds to a unique number.

[0086] S42. For any two bird trajectory segments, calculate the spatial distance between the two bird trajectory segments, the time label difference between the two bird trajectory segments, and the species label consistency parameter between the two bird trajectory segments by assigning spatial distance weight, time difference weight, and species consistency weight. Then, sum the above three values ​​according to the spatial distance weight, time difference weight, and species consistency weight to obtain the basic fitness parameter. The basic fitness parameter is used to measure the ecological splicing cost between any two bird trajectory segments. The spatial distance is measured by calculating the geographical distance between the end point of the first bird trajectory segment and the start point of the second bird trajectory segment. The time label difference is the absolute value of the end time label of the first bird trajectory segment and the start time label of the second bird trajectory segment. The species consistency parameter is zero when the species labels of the two bird trajectory segments are the same and one when they are different. The sum of the spatial distance weight, time difference weight, and species consistency weight is one.

[0087] S43. Based on the basic fitness parameters, wind direction matching parameters, terrain elevation constraint parameters, and historical migration record consistency parameters are introduced to reflect the matching of the angle between wind direction and flight direction between two bird trajectory segments, the terrain elevation jump between two bird trajectory segments, and the splicing consistency of two bird trajectory segments in historical data, respectively. By assigning wind direction matching correction weights, terrain elevation correction weights, and historical migration record correction weights, the wind direction matching parameters, terrain elevation constraint parameters, and historical migration record consistency parameters are added as correction terms to the basic fitness parameters to obtain enhanced fitness parameters. Enhanced fitness parameters are used to measure the comprehensive splicing cost of two bird trajectory segments under ecological and physical constraints. The closer the value of the wind direction matching parameter is to one, the more consistent the wind direction and flight direction are. The terrain elevation constraint parameter is used to determine whether the terrain elevation difference exceeds the range that the species can fly over. The historical migration record consistency parameter is used to determine whether the splicing instance exists in the historical data.

[0088] S44. The male lion performs a global search in the bird trajectory fragment dataset. For each geographic zone, all bird trajectory fragment pairs in the geographic zone are combined and matched. The enhancement fitness parameter of each bird trajectory fragment pair is calculated. The bird trajectory fragment pairs with the lowest enhancement fitness parameters are taken as the preliminary candidate pairing results. The preliminary candidate pairing results form the global candidate pairing set.

[0089] S45. Based on the global candidate pairing set, a migration ecological filtering mechanism is introduced by the female lion. For each candidate pairing, it is determined in turn whether its time tag interval is less than or equal to the maximum allowable migration time jump threshold, whether the wind direction matching parameter is greater than or equal to the lower limit of the wind-assisted effect, and whether the terrain climbing constraint parameter is equal to zero. Only when all conditions are met is the pairing retained as a locally effective pairing. All locally effective pairs form a local candidate pairing set.

[0090] S46. For unmatched bird trajectory segments in the bird trajectory segment dataset, introduce temporal and spatial perturbations to randomly perturb the time labels and spatial coordinates of the cubs to generate new combinations, and calculate the enhanced fitness parameter of each combination. Apply the same screening criteria as above to all new combinations, retain valid pairs, and form a perturbation expansion candidate set for all valid pairs.

[0091] S47. All valid pairs in the local candidate pairing set and the perturbation expansion candidate set are spliced ​​together in chronological order according to the time label to obtain a complete candidate trajectory. Each complete candidate trajectory consists of several sequentially spliced ​​bird trajectory segments. During the trajectory splicing process, the enhancement fitness parameters between each group of bird trajectory segments are recorded. The arithmetic mean of all enhancement fitness parameters is used as the average path splicing cost of the complete candidate trajectory.

[0092] S48. For each complete candidate trajectory, based on the path splicing average cost, the path splicing average cost is converted into a confidence parameter using a negative normalized exponential function. The standardization factor is used to adjust the distribution range of the confidence parameter. The confidence parameters of all complete candidate trajectories form a scoring matrix. All complete candidate trajectories and the scoring matrix are output to form a candidate trajectory dataset.

[0093] In this embodiment, S5 includes the following steps:

[0094] S51. The candidate trajectory dataset consists of several candidate trajectories. Each candidate trajectory is composed of several sequentially spliced ​​bird trajectory segments. Each bird trajectory segment contains spatial coordinates, time labels, and bird species labels. The candidate trajectory dataset is transformed into a sequence structure that the model can recognize, resulting in a set of candidate trajectory sequences. Each candidate trajectory sequence is a time-ordered sequence.

[0095] S52. For the candidate trajectory sequence set, calculate the spatial variation based on the longitude and latitude of each trajectory point, and convert the spatial variation into polar coordinates to obtain the geographic argument parameter and geographic modulus parameter. The geographic argument parameter is used to measure the spatial angular distance between two trajectory points, and the geographic modulus parameter is used to measure the geographic distance between two trajectory points. The geographic argument parameter and geographic modulus parameter, together with the time label difference, form a rotated geographic angular distance code. The rotated geographic angular distance code is used as the spatial and temporal feature embedding of the candidate trajectory sequence, and the rotated geographic angular distance code corresponds one-to-one with each trajectory point.

[0096] S53. The candidate trajectory sequence and its corresponding rotation geographic distance code are input into the spatiotemporal Transformer model configured with rotation position embedding. The spatiotemporal Transformer model directly uses the geographic argument parameter and geographic magnitude parameter of the rotation geographic distance code in each attention head to measure the spatial relationship between trajectory points, and combines the time label difference as a measure of temporal relationship. In the process of calculating attention weight, the attention distribution ratio between different trajectory points is dynamically adjusted based on the geographic argument parameter and geographic magnitude parameter of the rotation geographic distance code, so as to realize multi-dimensional collaborative perception of space and time, and output a multi-head attention weight matrix. All multi-head attention weight matrices correspond one-to-one with the candidate trajectory sequence.

[0097] S54. Based on the multi-head attention weight matrix, perform long-span sequence alignment for each candidate trajectory sequence. During the sequence alignment process, for each trajectory segment and its context segment, use the geographic argument parameter and geographic modulus parameter of the rotation geographic distance encoding to dynamically adjust the temporal arrangement order of trajectory points and optimize the spatial connection structure between trajectory points to ensure that the aligned trajectory sequence maintains continuity and rationality in both spatial and temporal dimensions, and generate an aligned trajectory sequence set.

[0098] S55. Based on the aligned trajectory sequence set, bird individual identification is performed based on the geographic argument parameter, geographic modulus parameter, and time tag difference of the rotation geographic distance encoding of each trajectory point. By comparing the geographic argument parameter, geographic modulus parameter, and time tag difference between different trajectory points in the same trajectory sequence, the spatial proximity, temporal continuity, and motion consistency of the trajectory points are comprehensively judged. Trajectory points belonging to the same bird are aggregated into a unified identity, resulting in a set of labeled trajectory identity tags. The set of trajectory identity tags corresponds one-to-one with the aligned trajectory sequence set.

[0099] S56. For the missing segment regions in the aligned trajectory sequence set, the geographic argument parameter and geographic modulus parameter of the rotated geographic angle distance encoding and their corresponding time tag difference are used. Based on the interpolation prediction method, the spatial coordinate prediction value and time tag prediction value of the missing segment are calculated. The prediction value is added to the missing segment region to realize the continuity repair of the trajectory sequence and generate a continuous and complete flight path set.

[0100] S57. All continuous and complete flight path sets are uniformly formatted. Each continuous and complete flight path contains temporally associated spatial coordinates, time labels, bird species labels, trajectory identity labels, and rotational geographic distance codes corresponding to each trajectory point. The continuous flight path dataset is then output.

[0101] In this embodiment, S6 includes the following steps:

[0102] S61. Write the continuous flight path dataset into the trajectory database. The trajectory database records the spatial coordinates, time label, bird species label, trajectory identity label, rotation geographic distance code and weight parameters of each flight path to form a stored continuous flight path dataset.

[0103] S62. Associate the stored continuous flight path dataset with the bird ecological knowledge graph. Based on the spatial coordinates, time labels and bird species labels of each flight path, extract the corresponding habitat type, migratory species, migration time window and ecological risk level parameters from the bird ecological knowledge graph to generate a path-knowledge graph matching parameter set.

[0104] S63. Associate the stored continuous flight path dataset with real-time environmental data. Based on the spatial coordinates, time labels and weight parameters of each flight path, extract the corresponding meteorological conditions, surface features and environmental disturbance parameters from the real-time environmental data to generate a path-environment parameter set.

[0105] S64. Based on the path-knowledge graph matching parameter set and the path-environment parameter set, output the cross-regional bird migration monitoring result set, and output the set of all cross-regional bird migration monitoring results.

[0106] In this embodiment, the cross-regional bird migration monitoring results set includes the following result types:

[0107] The habitat type identification results are determined based on the spatial coordinates of the flight path and the habitat type parameters in the knowledge graph, generating a habitat type label for each flight path;

[0108] Based on the identification results of high-risk migration areas, the spatiotemporal distribution monitoring results of species in the flight path and knowledge graph, and the bird species tags, time tags and spatial coordinates in the environmental data, the spatiotemporal distribution change trends of each species are output.

[0109] Based on the spatial overlap and temporal consistency parameters of multiple flight paths, the migration corridor identification results output migration corridor segments and corridor activity indicators.

[0110] A bird flight path intelligent tracking and monitoring system based on big data analysis, used to execute a bird flight path intelligent tracking and monitoring method based on big data analysis, includes the following modules:

[0111] The multimodal bird observation data acquisition module is used to acquire multimodal bird observation data, including images, videos, and sounds, in real time and generate raw bird observation datasets.

[0112] The data preprocessing module is used to perform data preprocessing on the original bird observation dataset, locate and identify bird targets in images and videos, identify bird species corresponding to sound signals, and output a standardized bird observation dataset.

[0113] The trajectory segment modeling module is used to establish the temporal trajectory of the same bird based on a standardized bird observation dataset, call a multi-target tracking algorithm, fuse multimodal recognition results, and construct a trajectory segment dataset.

[0114] The lion pride optimization module is used to perform lion pride optimization search on the trajectory segment dataset based on cross-regional migration constraints. It uses spatial distance weight, temporal difference weight, species consistency weight and ecological correction parameters to obtain candidate trajectory datasets and confidence score matrices.

[0115] The Spatiotemporal Transformer fine-grained alignment module is used to input candidate trajectory datasets into a spatiotemporal Transformer model configured with rotational geographic distance encoding. Based on the rotational geographic distance encoding, it performs long-span sequence alignment, bird individual identification, and missing segment interpolation to generate a continuous flight path dataset.

[0116] The trajectory evaluation and knowledge fusion module is used to evaluate the error of the continuous flight path dataset and update the weight parameters based on the evaluation results before writing them into the trajectory database. It associates the stored continuous flight path dataset with the bird ecological knowledge graph and real-time environmental data, and outputs a set of cross-regional bird migration monitoring results.

[0117] Example 1:

[0118] During continuous monitoring, the system received data samples uploaded from multiple observation terminals. One bird individual trajectory data segment was numbered "P000321". This data segment contained 52 time-series trajectory points. Each trajectory point recorded spatial coordinates ([119.028, 32.108]), a time label ("2023-03-15 10:23:01"), and a bird species label of "Species-A". The confidence level of each single-point observation was greater than 0.82. Two time intervals, 2.7 hours and 4.3 hours, were missing from the uploaded batch of data segments, indicating spatial drift. Traditional methods use the nearest neighbor principle to stitch P000321 with two other trajectory segments numbered "P000347" and "P000412". The final overall spatial error reached 2.16 kilometers, and some identity labels were assigned repeatedly, resulting in a decrease in trajectory integrity and individual consistency.

[0119] After adopting the method of this invention, the basic ecological constraint fitness assessment is first performed on the P000321 fragment and all uploaded trajectory fragments. The system calculates the weighted scores of P000321 and its candidate splicing fragments in terms of spatial distance, time tag difference, species tag consistency, wind direction parameter, and terrain ascent constraint. For the splicing attempt between P000321 and P000347, the spatial distance parameter is 0.97 km, the time tag difference is 1.8 hours, the species tags are consistent, the wind direction matching parameter is 0.87, the terrain ascent constraint parameter is 0, and the final enhanced fitness parameter is 0.52. For the splicing attempt between P000321 and P000412, the spatial distance parameter is 2.13 km, the time tag difference is 3.9 hours, the species are consistent, the wind direction matching parameter is 0.51, the terrain ascent constraint parameter is 0.23, and the enhanced fitness parameter is 1.48. The system prioritizes retaining P000347, which has the lowest fitness parameter, as the next segment for splicing, and automatically discards the splicing path of P000412.

[0120] Subsequently, to address the 2.7-hour observation gap in the spliced ​​sections P000321 and P000347, the system invoked the spatiotemporal Transformer module with rotated geographic angular distance encoding for interpolation repair. The interpolation module converted the spatial polar coordinate changes of the preceding and following trajectory points into geographic argument parameters (e.g., 62.7 degrees, 64.3 degrees), geographic modulus parameters (e.g., 1.22 km, 1.01 km), and time label differences (2.7 hours), dynamically adjusting the spatial position of the interpolation points in the sequence. After interpolation, three trajectory interpolation points were generated, with spatial errors of 0.92 km, 1.08 km, and 1.17 km for each point, lower than the average of 2.01 km for the traditional Transformer method.

[0121] In the identity verification stage, the system compares the complete trajectory with the identity feature vectors of historical migratory individuals in the database. The spatial trajectory similarity threshold is 0.91, and the species label consistency threshold is 1. Finally, P000321, P000347 and all trajectory points in the interpolation segment are uniformly labeled as the same bird individual "IDX_1876".

[0122] After the entire process of splicing and repair is completed, the system automatically evaluates the average spatial error of the continuous flight path, which is 1.07 kilometers, lower than the system's set upper limit of 2.0 kilometers. The trajectory confidence weight is increased to 0.93 and stored in the trajectory database.

[0123] The knowledge graph was synchronized with real-time environmental data. The system detected that the trajectory overlapped with meteorological parameters (wind speed exceeding 10.5 m / s) in a specific section (trajectory point numbers 27-41). The ecological knowledge graph marked this section as a potential high-risk migration area, and the system automatically output a high-risk area label in the monitoring results report, associating it with the risk factor of sudden wind speed changes. Simultaneously, based on historical data of similar individuals, the system determined that the habitat type continuously traversed by this trajectory was riverine wetlands-coastal mudflats, and recorded the species' typical migration routes for the current year.

[0124] Comparative experiments show that when processing this batch of data using the traditional nearest neighbor + absolute position encoding Transformer, the average spatial error of P000321 stitching is 2.23 km, the breakpoint interpolation error is 2.14 km, the number of misassigned identity tags is 2, the recall rate for habitat risk areas is 0.81, and the stitching time is 37.1 seconds. After processing using the method of this invention, the average spatial error of the same batch stitching is 1.07 km, the breakpoint interpolation error is 1.06 km, the number of misassigned identity tags is 0, the recall rate for risk areas is 0.93, and the stitching time is 19.2 seconds.

[0125] Another trajectory segment, numbered P001765, failed to form a complete cross-regional path after being stitched together using traditional methods due to mis-stitching, resulting in misalignment of the overall time series and switching of individual identities. However, the method of this invention automatically eliminated unreasonable paths, ultimately generating a high-confidence path spanning four geographical zones and covering 78 continuous trajectory points, with a mean spatial error of 1.22 kilometers and 100% identity consistency.

[0126] In summary, throughout the entire continuous monitoring process, the method of this invention can automatically complete the normalization, fragment splicing, missing data repair, and risk zone identification of bird trajectory data under massive, multi-source, and complex conditions. The core indicators are significantly better than traditional methods. All data in the process are recorded in the database, making them traceable and verifiable. This proves that the invention not only improves splicing accuracy, interpolation precision, and identity consistency, but also greatly enhances the ability to identify ecological risks and monitor them in real time.

[0127] In the lion pride optimization search model, this invention combines geographical zone zoning, wind direction matching, terrain ascent constraints, and ecological characteristics of historical migration record consistency to impose multi-level constraints and corrections on the global pairing and local screening of trajectory fragments. This enables intelligent coarse concatenation of trajectory fragments across multiple targets and regions. Compared with traditional splicing methods based on heuristic thresholds or Euclidean distance, this invention can dynamically adapt to the migration behavior of different species, geographical environments, and climatic conditions, effectively reducing the probability of mis-sponging and missed splicing.

[0128] This invention proposes a rotational geographic distance encoding method that combines spatial argument parameters and geographic modulus parameters with time label differences. This method is embedded in a multi-head attention mechanism to achieve dynamic correlation in the spatial and temporal dimensions of trajectory sequences. It can accurately model bird flight paths under long-distance and complex geographic conditions. The Transformer can capture long-distance dependence and periodic migration patterns in sequence alignment, individual identification, and missing segment interpolation, achieving reconstructed results with controllable errors and coherent trajectories.

[0129] This invention combines multi-source observation data (images, sound, trajectories, and meteorological data) with a dynamic weight update mechanism, and integrates knowledge graph association with environmental data fusion. It can output real-time results for bird habitat type identification, high-risk migration zone identification, species spatiotemporal distribution monitoring, and multi-dimensional monitoring of migration corridor activity. In actual deployment, it exhibits extremely high adaptability and scalability, and can cover the continuous tracking needs of bird migration under different time periods, multiple geographical zones, and heterogeneous equipment collection conditions. It effectively supports application scenarios such as ecological protection, migration research, and airspace management, and has significant engineering promotion and ecological protection value.

[0130] The above description is only a preferred embodiment 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 intelligent tracking and monitoring of bird flight paths based on big data analysis, characterized in that, Includes the following steps: S1. Real-time acquisition of multimodal bird observation data including images, videos, and sounds to generate raw bird observation datasets; S2. Perform data preprocessing on the original bird observation dataset to output a standardized bird observation dataset; S3. Based on a standardized bird observation dataset, a multi-target tracking algorithm is called to establish the temporal trajectory of the same bird in consecutive video frames, thus obtaining a bird trajectory fragment dataset; S4. Construct a lion pride optimization search model by inputting the bird trajectory fragment dataset into the lion pride optimization search model to form a candidate trajectory dataset; S4 includes the following steps: S41. Construct a lion pride optimization search model based on cross-regional migration constraints, taking the bird trajectory fragment dataset as input. The bird trajectory fragment dataset is divided into geographical zones, and each geographical zone corresponds to a unique number. S42. For any two bird trajectory segments, calculate the spatial distance between the two bird trajectory segments, the time label difference between the two bird trajectory segments, and the consistency parameter between the species labels of the two bird trajectory segments by assigning spatial distance weight, time difference weight, and species consistency weight. Then, sum the above three values ​​according to the spatial distance weight, time difference weight, and species consistency weight to obtain the basic fitness parameter. S43. Based on the basic fitness parameters, by separately allocating wind direction matching correction weight, terrain ascent correction weight, and historical migration record correction weight, the wind direction matching parameter, terrain ascent constraint parameter, and historical migration record consistency parameter are added as correction terms to the basic fitness parameters to obtain the enhanced fitness parameters. S44. The male lion performs a global search in the bird trajectory fragment dataset. For each geographic zone, all bird trajectory fragment pairs in the geographic zone are combined and matched. The enhancement fitness parameter of each bird trajectory fragment pair is calculated. The bird trajectory fragment pairs with the lowest enhancement fitness parameters are taken as the preliminary candidate pairing results. The preliminary candidate pairing results form the global candidate pairing set. S45. Based on the global candidate pairing set, a migration ecological filtering mechanism is introduced for each candidate pairing. For each candidate pairing, it is determined whether its time tag interval is less than or equal to the maximum allowable migration time jump threshold, whether the wind direction matching parameter is greater than or equal to the lower limit of the wind-assisted effect, and whether the terrain climbing constraint parameter is equal to zero. Only when all conditions are met is the pairing retained as a locally effective pairing. All locally effective pairings form a local candidate pairing set. S46. For unmatched bird trajectory segments in the bird trajectory segment dataset, introduce temporal and spatial perturbations to randomly perturb the time labels and spatial coordinates of the cubs to generate new combinations, and calculate the enhanced fitness parameter of each combination. Apply the same screening criteria as above to all new combinations, retain valid pairs, and form a perturbation expansion candidate set for all valid pairs. S47. All valid pairs in the local candidate pairing set and the perturbation expansion candidate set are spliced ​​together in chronological order according to the time label to obtain a complete candidate trajectory. During the trajectory splicing process, the enhancement fitness parameters between each group of bird trajectory segments are recorded. The arithmetic mean of all enhancement fitness parameters is used as the average cost of path splicing for the complete candidate trajectory. S48. For each complete candidate trajectory, based on the average cost of path splicing, the average cost of path splicing is transformed into a confidence parameter using a negative normalized exponential function. The confidence parameters of all complete candidate trajectories form a scoring matrix, thus forming a candidate trajectory dataset. S5. Input the candidate trajectory dataset into the spatiotemporal Transformer model configured with rotational position embedding, perform long-span sequence alignment, bird individual identification and missing segment interpolation based on rotational geographic angular distance encoding, and obtain a continuous and complete set of flight paths to generate a continuous flight path dataset; S5 includes the following steps: S51. The candidate trajectory dataset consists of several candidate trajectories. The candidate trajectory dataset is transformed into a sequence structure that the model can recognize, resulting in a set of candidate trajectory sequences. S52. For the candidate trajectory sequence set, calculate the spatial change based on the longitude and latitude of each trajectory point, and convert the spatial change into polar coordinates to obtain the geographic argument parameter and geographic modulus parameter, and combine them with the time tag difference to form the rotated geographic angular distance code. S53. The candidate trajectory sequence and its corresponding rotation geographic distance code are input into the spatiotemporal Transformer model configured with rotation position embedding. In each attention head, the spatiotemporal Transformer model directly uses the geographic argument parameter and geographic magnitude parameter of the rotation geographic distance code to measure the spatial relationship between trajectory points, and combines the time label difference as a measure of temporal relationship. In the process of calculating attention weight, the attention allocation ratio between different trajectory points is dynamically adjusted based on the geographic argument parameter and geographic magnitude parameter of the rotation geographic distance code, and the multi-head attention weight matrix is ​​output. S54. Based on the multi-head attention weight matrix, perform long-span sequence alignment for each candidate trajectory sequence. During the sequence alignment process, for each trajectory segment and its context segment, dynamically adjust the temporal arrangement order of trajectory points using the geographic argument parameter and geographic modulus parameter of the rotation geographic distance encoding, and optimize the spatial connection structure between trajectory points to generate an aligned trajectory sequence set. S55. Based on the aligned trajectory sequence set, bird individual identification is performed. By comparing the geographic argument parameter, geographic modulus parameter and time tag difference between different trajectory points in the same trajectory sequence, the spatial proximity, temporal continuity and motion consistency of the trajectory points are comprehensively judged. The trajectory points belonging to the same bird are aggregated into a unified identity to obtain a set of labeled trajectory identity tags. S56. For the missing trajectory segment regions in the aligned trajectory sequence set, the geographic argument parameter and geographic modulus parameter of the rotated geographic distance encoding and their corresponding time label difference are used. Based on the interpolation prediction method, the spatial coordinate prediction value and time label prediction value of the missing segment are calculated. The prediction values ​​are then added to the missing segment regions to generate a continuous and complete flight path set. S57. Perform uniform formatting on all continuous and complete flight path sets, and output a continuous flight path dataset; S6. Based on the continuous flight path dataset, a stored continuous flight path dataset is formed. The stored continuous flight path dataset is then associated with the bird ecological knowledge graph and real-time environmental data to output a set of cross-regional bird migration monitoring results.

2. The intelligent tracking and monitoring method for bird flight paths based on big data analysis according to claim 1, characterized in that, S2 includes the following steps: S21. Perform bird target detection and recognition processing on the image and video data in the original bird observation dataset, and output the set of spatial coordinates of all detected bird targets in each frame of image or video. S22. Use an audio recognition model to identify bird species from the sound data in the original bird observation dataset, extract the voiceprint features of each sound data segment, and match the voiceprint features with known bird species labels to generate an audio recognition result set; S23. Based on the GPS labels in the original bird observation dataset, perform time normalization and spatial coordinate normalization on all observation data to obtain a standardized observation data set; S24. Perform confidence screening and duplicate observation removal on the standardized observation dataset to obtain the confidence screening set; S25. Perform outlier cleaning on the confidence screening set, calculate the Euclidean distance between each observation data and the mean vector in the confidence screening set, retain only the observation data whose Euclidean distance is not higher than the outlier discrimination threshold, and finally output the standardized bird observation dataset.

3. The intelligent tracking and monitoring method for bird flight paths based on big data analysis according to claim 2, characterized in that, S3 includes the following steps: S31. For the standardized bird observation dataset, a multi-target tracking algorithm is used to assign the identity of each bird target in continuous video frames. In each frame, each spatial coordinate in the spatial coordinate set is processed according to the horizontal coordinate, vertical coordinate and time label. By comparing the correlation between the spatial coordinates and time labels of adjacent frames, the spatial coordinates of the same bird in different frames are associated to form an initial time-series trajectory set. S32. Extract the bird species label and time label corresponding to each sound data segment from the audio recognition result set in the standardized bird observation dataset. Match and associate the bird species label identified under the same time label with the spatial coordinates matched in the video frame. Fuse the image, video and sound recognition results to construct a fused observation segment set containing spatial coordinates, time labels and bird species labels. S33. Based on the trajectory similarity index function, the initial time-series trajectory set and the fused observation segment set are fused to obtain the trajectory segment set; S34. Perform uniform formatting on all trajectory segments in the trajectory segment set so that each trajectory segment contains temporally associated spatial coordinates, time labels, and bird species labels, and output the bird trajectory segment dataset.

4. The intelligent tracking and monitoring method for bird flight paths based on big data analysis according to claim 3, characterized in that, The trajectory similarity index function value is obtained by calculating the Euclidean distance between the initial time-series trajectory and the fused observation segment in spatial coordinates, calculating the consistency measure between the initial time-series trajectory and the fused observation segment in species label, multiplying the Euclidean distance and the consistency measure by the corresponding normalized weight coefficients and summing them, and obtaining the value by taking the reciprocal of the sum and adding one. By maximizing the trajectory similarity index function value, the trajectories and observation segments of the same bird are fused to obtain a set of trajectory segments.

5. The intelligent tracking and monitoring method for bird flight paths based on big data analysis according to claim 1, characterized in that, S6 includes the following steps: S61. Write the continuous flight path dataset into the trajectory database. The trajectory database records the spatial coordinates, time label, bird species label, trajectory identity label, rotation geographic distance code and weight parameters of each flight path to form a stored continuous flight path dataset. S62. Associate the stored continuous flight path dataset with the bird ecological knowledge graph. Based on the spatial coordinates, time labels and bird species labels of each flight path, extract the corresponding habitat type, migratory species, migration time window and ecological risk level parameters from the bird ecological knowledge graph to generate a path-knowledge graph matching parameter set. S63. Associate the stored continuous flight path dataset with real-time environmental data. Based on the spatial coordinates, time labels and weight parameters of each flight path, extract the corresponding meteorological conditions, surface features and environmental disturbance parameters from the real-time environmental data to generate a path-environment parameter set. S64. Based on the path-knowledge graph matching parameter set and the path-environment parameter set, output the cross-regional bird migration monitoring result set, and output the set of all cross-regional bird migration monitoring results.

6. The intelligent tracking and monitoring method for bird flight paths based on big data analysis according to claim 5, characterized in that, The collection of cross-regional bird migration monitoring results includes the following result types: The habitat type identification results are determined based on the spatial coordinates of the flight path and the habitat type parameters in the knowledge graph, generating a habitat type label for each flight path; Based on the identification results of high-risk migration areas, the spatiotemporal distribution monitoring results of species in the flight path and knowledge graph, and the bird species tags, time tags and spatial coordinates in the environmental data, the spatiotemporal distribution change trends of each species are output. Based on the spatial overlap and temporal consistency parameters of multiple flight paths, the migration corridor identification results output migration corridor segments and corridor activity indicators.

7. A bird flight path intelligent tracking and monitoring system based on big data analysis, used to execute the bird flight path intelligent tracking and monitoring method based on big data analysis as described in any one of claims 1-6, characterized in that, Includes the following modules: The multimodal bird observation data acquisition module is used to acquire multimodal bird observation data, including images, videos, and sounds, in real time and generate raw bird observation datasets. The data preprocessing module is used to perform data preprocessing on the original bird observation dataset, locate and identify bird targets in images and videos, identify bird species corresponding to sound signals, and output a standardized bird observation dataset. The trajectory segment modeling module is used to establish the temporal trajectory of the same bird based on a standardized bird observation dataset, call a multi-target tracking algorithm, fuse multimodal recognition results, and construct a trajectory segment dataset. The lion pride optimization module is used to perform lion pride optimization search on the trajectory segment dataset based on cross-regional migration constraints. It uses spatial distance weight, temporal difference weight, species consistency weight and ecological correction parameters to obtain candidate trajectory datasets and confidence score matrices. The Spatiotemporal Transformer fine-grained alignment module is used to input candidate trajectory datasets into a spatiotemporal Transformer model configured with rotational geographic offset encoding. Based on the rotational geographic offset encoding, it performs long-span sequence alignment, bird individual identification, and missing segment interpolation to generate a continuous flight path dataset. The trajectory evaluation and knowledge fusion module is used to evaluate the error of the continuous flight path dataset and update the weight parameters based on the evaluation results before writing them into the trajectory database. It associates the stored continuous flight path dataset with the bird ecological knowledge graph and real-time environmental data, and outputs a set of cross-regional bird migration monitoring results.

Images