A migratory bird tracking habitat identification method based on spatiotemporal fusion and adaptive kernel density

Abstract

This invention discloses a method for migratory bird tracking and habitat identification based on spatiotemporal fusion and adaptive kernel density. The method includes: processing sampled migratory bird trajectory data; defining temporal locality indices, duration potential, spatial dispersion indices, and spatial heterogeneity indices based on the processed data; weighted fusion of the defined data to obtain a comprehensive spatiotemporal resting confidence function, comparing it with a set of preset thresholds to obtain a set of high-confidence resting points for individual migratory birds, and performing cluster analysis to obtain preliminary clusters. After saliency filtering, a core habitat cluster set is output. Based on this, feature vectors for the clusters are constructed and calculated to obtain an inter-cluster dissimilarity matrix. Spatial clustering fusion is performed on the inter-cluster dissimilarity matrix to obtain a fused super-cluster set, which serves as the flight path for the migratory bird trajectory data. This invention, through hierarchical clustering and filtering mechanisms, can improve the ecological rationality and spatial coherence of habitat identification results.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary design field of ecological protection, animal movement ecology and spatiotemporal data science, and specifically involves a migratory bird tracking habitat identification method based on spatiotemporal fusion and adaptive kernel density. Background Technology

[0002] In recent years, the widespread application of lightweight satellite / GPS trackers and remote sensing environmental data has enabled researchers to acquire a large amount of time-series data on migratory bird locations to identify migration routes and stopover points. However, traditional methods typically use linear or standard spline interpolation to fill in missing parts of the trajectory, failing to fully consider the spatiotemporal complexity of migratory bird behavior. Especially when signals are interrupted for extended periods or behavioral patterns are varied (such as circling or gliding), the interpolation results often deviate from the actual movement trajectory, thus affecting the accuracy of subsequent analysis. When identifying stopover points, most methods rely on pre-set global thresholds (such as the duration of stay within a fixed radius) for judgment, lacking fusion analysis of local behavioral context and multimodal features. This easily leads to misjudging short-term dwelling, loitering, or circling behaviors as stopovers, resulting in limited identification accuracy. Furthermore, common clustering methods struggle to simultaneously and effectively capture both high-density, small-scale habitats and low-density, large-scale habitat areas. Clustering results often exhibit excessive spatial fragmentation or insufficient merging, failing to accurately reflect the actual habitat utilization pattern. More notably, existing methods often lack deep integration with ecological rules and fail to further validate and screen clustering results. This results in the final output of habitat areas containing a large amount of invalid or ecologically ambiguous noise, limiting their application value in practical conservation decision-making. Overall, traditional methods suffer from low automation and insufficient scalability when processing multi-source, heterogeneous, and large-scale satellite tracking data, making it difficult to meet the growing demands for data processing and ecological interpretation.

[0003] This invention aims to solve the following technical problems: 1. To address the uneven sampling intervals and fluctuating positioning accuracy of trajectory data, an adaptive and behavior-aware trajectory reconstruction and interpolation strategy is proposed, enabling the recovery of biologically reliable continuous trajectories in short, long, or behaviorally complex segments, and providing a quantifiable confidence assessment of the interpolation results; 2. To address the problem of misjudgment or missed judgment in traditional single-threshold pause determination, a pause confidence discrimination model integrating multi-dimensional spatiotemporal features such as temporal locality, duration, spatial dispersion, and spatial heterogeneity is constructed, thereby significantly improving the discrimination accuracy and robustness of pause points and pause events; 3. To address the issues of fixed parameter density estimation and clustering... To address the limitations of existing methods in multi-scale and multi-density scenarios, this invention proposes a multi-scale clustering process that combines kernel density estimation based on point-level adaptive bandwidth with density peak detection. This process attracts and assigns point clusters through density gradients to automatically identify habitat core areas with different intensities and non-convex shapes. Furthermore, it eliminates spurious clusters through cluster salience scoring to reduce fragmentation and missed detections. Fourthly, to further improve the ecological rationality and spatial coherence of the results, this invention also integrates hierarchical clustering and ecological rules to fuse and filter the initially identified habitat core areas, merge spatially adjacent clusters with similar attributes, and eliminate abnormal areas based on rules such as spatial isolation. Finally, it outputs the distribution of migratory bird habitats with clear ecological significance. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a migratory bird tracking habitat identification method based on spatiotemporal fusion and adaptive kernel density, aiming to solve the problems in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for migratory bird tracking and habitat identification based on spatiotemporal fusion and adaptive kernel density, comprising:

[0006] Step S1: The sampled migratory bird trajectory data is standardized. The standardized migratory bird trajectory data is then reconstructed using a three-level interpolation strategy to obtain the pre-processed and adaptively interpolated migratory bird trajectory data.

[0007] Step S2: Based on the migratory bird trajectory data after data preprocessing and adaptive interpolation, define the temporal locality index, duration potential, spatial dispersion index, and spatial heterogeneity index; weight and fuse the temporal locality index, duration potential, spatial dispersion index, and spatial heterogeneity index to obtain a comprehensive spatiotemporal rest confidence function; compare the comprehensive spatiotemporal rest confidence function with the set rest judgment threshold to obtain the set of high-confidence rest points for individual migratory birds;

[0008] Step S3: Perform cluster analysis on the set of high-confidence stopover points of migratory birds to obtain preliminary clusters. Perform saliency filtering on the preliminary clusters and output the set of core habitat clusters.

[0009] Step S4: Based on the cluster set of the output habitat core area, construct the feature vector of the cluster, calculate the cluster dissimilarity matrix by weighted metric on the cluster feature vector, and perform spatial clustering fusion on the cluster dissimilarity matrix to obtain the fused super cluster set, which serves as the flight path of migratory bird trajectory data.

[0010] Furthermore, the specific process of step S1 is as follows:

[0011] The sampled migratory bird trajectory data is imported into a unified data framework, and the migratory bird trajectory data is represented as follows:

[0012] ;

[0013] In the formula, This is the timestamp of the i-th migratory bird at the k-th trajectory point; Let x be the x-coordinate of the i-th migratory bird at the k-th trajectory point; Let be the ordinate of the i-th migratory bird at the k-th trajectory point; Let be the height of the i-th migratory bird at the k-th trajectory point; Let be the positioning quality index of the i-th migratory bird at the k-th trajectory point; Let be the total number of points on the trajectory of the i-th migratory bird;

[0014] Unifying irregularly sampled migratory bird trajectory data to a fixed time resolution, which is the target step size. ;

[0015] To calculate the estimated position of migratory birds at time t of the point to be interpolated, the width of the time window is first adaptively determined based on the sampling interval distribution of the migratory bird trajectory data. and set the time window as Within the time window There are 1 original observation point, and the timestamp of the original observation point is 1. The corresponding position is ; ,based on calculate ,express:

[0016] ;

[0017] In the formula, Estimate the position of migratory birds at time t, the interpolation point; For the first The weights of the original observation points; For the first The original observation points at the timestamp The corresponding actual location;

[0018] in, express:

[0019] ;

[0020] In the formula, It is a very small positive number; This is the absolute value of the time difference;

[0021] The specific process of using the three-stage interpolation strategy is as follows:

[0022] Let the two endpoints of the interpolation interval for the migratory bird's trajectory be respectively... and When the interval between the two endpoints of the interpolation interval satisfy This indicates that migratory bird trajectory data is missing; The timestamp of the left endpoint of the interpolation interval; for The actual location of migratory birds at all times; The timestamp of the right endpoint of the interpolation interval; for The actual location of migratory birds at all times; The time interval between the two endpoints of the interpolation interval;

[0023] When migratory bird trajectory data is missing, cubic Hermit splines with endpoint velocity constraints are used for interpolation reconstruction. Construction representation:

[0024] ;

[0025] In the formula, The basis function for the left endpoint position; The basis function is the position of the right endpoint; Let the left endpoint velocity basis function be used. The right-endpoint velocity basis function; left endpoint of the interpolation interval The instantaneous velocity vector of migratory birds at any given moment; The right endpoint of the interpolation interval The instantaneous velocity vector of migratory birds at any given moment; This is the normalized time parameter;

[0026] The method of sampling cubic Hermit splines ensures that in and Satisfying It conforms to the dynamic characteristics of short-term movement of migratory birds; for Estimated location of migratory birds at all times; for Estimated location of migratory birds at all times; for The instantaneous velocity vector of migratory birds at any given moment; for The instantaneous velocity vector of migratory birds at any given moment;

[0027] when At that time, neighborhood weighted interpolation was used. Based on the introduction of behavioral similarity adjustment, we obtain ,express:

[0028] ;

[0029] In the formula, To introduce interpolation weights for the j-th original observation point after behavioral similarity adjustment; The weight of the j-th original observation point; It is a natural exponential function; To measure and The differences between them; The motion feature vector of the local window where the j-th historical observation point is located; The motion feature vector of the local window where the interpolation point is located at time t; Here is the variance parameter;

[0030] when This indicates that the missing segments of migratory bird trajectory data are extremely long and lack reliable references. Therefore, a Monte Carlo candidate trajectory generation strategy is adopted: learning movement patterns from historical complete migratory bird trajectory data to generate... Article from arrive candidate trajectories ; For the first Candidate trajectories; The total number of candidate trajectories generated; The sign for "much greater than"; the range of "much greater than" is limited to: ,Right now ;

[0031] Calculate a confidence score, representing:

[0032] ;

[0033] In the formula, For the first Candidate trajectories Confidence score; Let m be the velocity of the m-th interpolation point on the candidate trajectory; These are velocity-type motion parameters estimated from migratory bird trajectory data; In order to be in Down, speed Probability distribution of occurrence; Let m be the angular velocity of the m-th interpolation point on the candidate trajectory; Angular velocity-type motion parameters estimated from migratory bird trajectory data; In order to be in Down, The probability distribution of angular velocity occurrence; Let be the vertical velocity of the m-th interpolation point on the candidate trajectory; Vertical velocity-type motion parameters estimated from migratory bird trajectory data; In order to be in Downward, vertical velocity Probability distribution of occurrence;

[0034] ;right arrive Compare and select the candidate trajectory with the highest score. ,Will As the interpolation result of the missing segment, i.e. ; Estimate the position of the interpolation point at time t; This is the interpolation result for the missing segment at time t.

[0035] Furthermore, the specific process for defining the temporal locality index, the duration potential, the spatial dispersion index, and the spatial heterogeneity index is as follows:

[0036] Each data point in the migratory bird trajectory data after data preprocessing and adaptive interpolation To define the temporal locality index, the duration potential, the spatial dispersion index, and the spatial heterogeneity index; For the first migratory bird trajectory data after data preprocessing and adaptive interpolation There are 10 data points; the data points include interpolation points and original observation points.

[0037] Define a temporal locality index, which represents:

[0038] ;

[0039] In the formula, for Temporal locality indicators; for Inner Time Dimension Standard deviation; for Inner Time Dimension The mean; Here, is the regularization parameter; ; For the first The time window corresponding to each data point; For the first The timestamps corresponding to each data point; The time window is half the width;

[0040] Define duration potential, representing:

[0041] ;

[0042] In the formula, for The potential for duration; This represents the number of neighboring points; For the first Timestamps of each data point; This is the duration decay coefficient;

[0043] Define the spatial scattering index, which represents:

[0044] ;

[0045] In the formula, for Spatial dispersion index; For the same time window The standard deviation of the longitude and latitude coordinates of all points within the plane; Spatial scale parameter;

[0046] Define a spatial heterogeneity index, which represents:

[0047] ;

[0048] In the formula, for Spatial heterogeneity index; In order to be in The number of points within the spatial neighborhood; for with neighbors Spatial distance between them; For the first migratory bird trajectory data after data preprocessing and adaptive interpolation One data point; This is the average distance; It is a very small positive number.

[0049] Furthermore, the specific process for obtaining the comprehensive spatiotemporal pause confidence function is as follows:

[0050] The comprehensive spatiotemporal pause confidence function is obtained, which represents:

[0051] ;

[0052] In the formula, for The comprehensive spatiotemporal pause confidence function; , , , They are respectively , , , Weighting coefficients; To obtain The minimum value between 1 and 1; These are the statistical characteristics of spatial distribution.

[0053] Furthermore, the specific process for obtaining the set of high-confidence resting points for individual migratory birds is as follows:

[0054] Set a global pause threshold ;when At that time, Determined as a rest stop ; ; For the first A rest stop; based on Obtain the set of high-confidence resting points for individual migratory birds. ; This includes the estimated resting points for all decisions.

[0055] Furthermore, the specific process for obtaining the preliminary clustering is as follows:

[0056] right Perform spatial kernel density estimation, spatial location ,express:

[0057] ;

[0058] In the formula, For spatial location The kernel density estimate at that location; for The total estimated number of stopping points; For the first The adaptive bandwidth corresponding to each estimated resting point; for The first in An estimated resting point; For standardized kernel functions;

[0059] get after, In spatial location The second-order partial derivatives are used to construct the density peak point matrix;

[0060] Finding through numerical optimization algorithms Density peak points where the gradient is zero, i.e. The density peak points represent the points where the gradient is zero. The density peak point matrix is ​​constructed from the second-order partial derivatives based on the density peak points where the gradient is zero. When the density peak point matrix is ​​negative definite, the density peak points are confirmed as local density maxima. Local density maxima are recorded as candidate cluster centers.

[0061] Will Assigning candidates to cluster centers using a density attraction-based allocation strategy. Iteratively execute the following gradient ascent process to find The convergent density peak point indicates:

[0062] ;

[0063] In the formula, For the first The estimated spatial location of the resting point after the next iteration; For the first The estimated spatial location of the resting point after the next iteration; This is the iteration step size; for exist The gradient vector at that point; Let be the norm of the gradient vector;

[0064] right The iterative process continues until... If the value is less than the set minimum threshold, stop the iteration. If the convergence points reach a single density peak, then the convergent density peak points are assigned to the same cluster. Preliminary clustering was completed; For the first Clusters.

[0065] Furthermore, the specific process of outputting the core habitat cluster set is as follows:

[0066] Saliency filtering is performed on the initial clustering to determine the significance of each cluster. Calculate significance score ,express:

[0067] ;

[0068] In the formula, For clusters The significance score; For the first Cluster The number of density peak points contained therein; For all clusters The average number of density peak points; For clusters The maximum value of the kernel density estimate for all density peak points within the range; The standard deviation of the kernel density estimate; For clusters The minimum value of the kernel density estimate for all density peak points within the range;

[0069] Set a significance threshold ,when At that time, retain cluster;

[0070] The retained clusters are processed using the α-shape algorithm. Perform calculations using the set parameters. Based on the estimated spatial distribution of resting points, a suitable polygonal boundary is generated. The polygonal boundary determines the core habitat clusters, and the set of core habitat clusters is output. ; For the first A collection of core habitat clusters; For the first A cluster of core habitat areas, The core control parameters of the α-shape algorithm are obtained through... Adjust the tightness of the spatial distribution of polygon boundaries and rest points.

[0071] Furthermore, the specific process for obtaining the inter-cluster dissimilarity matrix is ​​as follows:

[0072] by Using the basic unit, construct the cluster's feature vector, representing:

[0073] ;

[0074] In the formula, For the first The feature vectors of each cluster; For the first The longitude coordinates of each cluster; For the first The weighted center latitude coordinates of each cluster; For the first The area of ​​each cluster; For the first The total number of rest points contained in a cluster; For the first The average stopping confidence of all stopping points within a cluster; For the first The sum of the total dwell times of all resting points within a cluster;

[0075] Measure any two clusters and The dissimilarity and distance between them are calculated using a weighted metric that combines geographic distance and attribute distance, expressed as follows:

[0076] ;

[0077] In the formula, For clusters with cluster The overall dissimilarity between them; This represents the weighting coefficient for geographical distance; For clusters with cluster Geographical distance between center points; This is the normalization factor for geographical distance; This is the weighting coefficient for attribute distance; for and The distance between the attributes; For the first A collection of core habitat clusters; For the first The feature vectors of each cluster;

[0078] Will Arranged in row-column form, they form a cluster dissimilarity matrix.

[0079] Furthermore, the specific process of obtaining the fused supercluster set is as follows:

[0080] The average linking method in hierarchical clustering algorithm is used to analyze the clusters in the inter-cluster dissimilarity matrix. with cluster Perform cluster fusion to obtain a fused cluster set;

[0081] Set fusion distance threshold To merge all the clusters into one cluster with cluster The distance between them is less than The clusters are merged to obtain a supercluster set. ; This is the e-th supercluster after merging;

[0082] For each supercluster in the merged supercluster set First find the matching Another recent supercluster ,calculate and Distance between the centroids ;

[0083] Set the maximum isolation distance threshold ,when At that time, Remove; when At that time, the merged supercluster set Flight paths used as data for migratory bird trajectories.

[0084] Compared with existing technologies, the present invention has the following advantages:

[0085] (1) The present invention significantly enhances the scene adaptability and biological rationality of interpolation by using cubic Hermit splines with endpoint velocity constraints, neighborhood weighted interpolation in complex behavior, and Monte Carlo candidate trajectories with confidence-based selection in the case of extremely long missing segments of migratory bird trajectory data and lack of reliable references.

[0086] (2) The present invention uses a hierarchical clustering algorithm to automatically adjust the smoothing parameters according to the local density of the data, and accurately identify the core areas of habitats of different sizes, densities and forms.

[0087] (3) This invention establishes a hierarchical clustering and screening mechanism: combining hierarchical clustering with ecological rules, constructing an inter-cluster dissimilarity matrix, using average linking for cluster fusion, and introducing ecological rules such as spatial isolation test, effectively eliminating abnormal clusters and improving the ecological rationality and spatial coherence of habitat identification results. Attached Figure Description

[0088] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0089] like Figure 1 As shown, the present invention provides a technical solution: a migratory bird tracking habitat identification method based on spatiotemporal fusion and adaptive kernel density, comprising:

[0090] Step S1: The sampled migratory bird trajectory data is standardized. The standardized migratory bird trajectory data is then reconstructed using a three-level interpolation strategy to obtain the pre-processed and adaptively interpolated migratory bird trajectory data.

[0091] The sampled migratory bird trajectory data is imported into a unified data framework, and the migratory bird trajectory data is represented as follows:

[0092] ;

[0093] In the formula, This is the timestamp of the i-th migratory bird at the k-th trajectory point; Let x be the x-coordinate of the i-th migratory bird at the k-th trajectory point; Let be the ordinate of the i-th migratory bird at the k-th trajectory point; Let be the height of the i-th migratory bird at the k-th trajectory point; Let be the positioning quality index of the i-th migratory bird at the k-th trajectory point; Let be the total number of points on the trajectory of the i-th migratory bird;

[0094] Unifying irregularly sampled migratory bird trajectory data to a fixed time resolution, which is the target step size. ;

[0095] To calculate the estimated position of migratory birds at time t of the point to be interpolated, the width of the time window is first adaptively determined based on the sampling interval distribution of the migratory bird trajectory data. and set the time window as Within the time window There are 1 original observation point, and the timestamp of the original observation point is 1. The corresponding position is ; ,based on calculate ,express:

[0096] ;

[0097] In the formula, Estimate the position of migratory birds at time t, the interpolation point; For the first The weights of the original observation points; For the first The original observation points at the timestamp The corresponding actual location;

[0098] in, express:

[0099] ;

[0100] In the formula, It is a very small positive number; This is the absolute value of the time difference;

[0101] The specific process of using the three-stage interpolation strategy is as follows:

[0102] Let the two endpoints of the interpolation interval for the migratory bird's trajectory be respectively... and When the interval between the two endpoints of the interpolation interval satisfy This indicates that the data on migratory bird tracks is relatively short; The timestamp of the left endpoint of the interpolation interval; for The actual location of migratory birds at all times; The timestamp of the right endpoint of the interpolation interval; for The actual location of migratory birds at all times; The time interval between the two endpoints of the interpolation interval;

[0103] When the data indicating a short gap in migratory bird trajectory is presented, cubic Hermit splines with endpoint velocity constraints are used for interpolation reconstruction. Construction representation:

[0104] ;

[0105] In the formula, The basis function for the left endpoint position; The basis function is the position of the right endpoint; Let the left endpoint velocity basis function be used. The right-endpoint velocity basis function; left endpoint of the interpolation interval The instantaneous velocity vector of migratory birds at any given moment; The right endpoint of the interpolation interval The instantaneous velocity vector of migratory birds at any given moment; This is the normalized time parameter;

[0106] Wherein, the cubic Hermitian spline basis functions represent:

[0107] ;

[0108] The method of sampling cubic Hermit splines ensures that in and Satisfying It conforms to the dynamic characteristics of short-term movement of migratory birds; for Estimated location of migratory birds at all times; for Estimated location of migratory birds at all times; for The instantaneous velocity vector of migratory birds at any given moment; for The instantaneous velocity vector of migratory birds at any given moment;

[0109] when When the behavior is complex (e.g., there may be hovering), neighborhood-weighted interpolation is used. Based on the introduction of behavioral similarity adjustment, we obtain ,express:

[0110] ;

[0111] In the formula, To introduce interpolation weights for the j-th original observation point after behavioral similarity adjustment; The weight of the j-th original observation point; It is a natural exponential function; To measure and The differences between them; The motion feature vector of the local window where the j-th historical observation point is located; The motion feature vector of the local window where the interpolation point is located at time t; Here is the variance parameter;

[0112] when This indicates that the missing segments of migratory bird trajectory data are extremely long and lack reliable references. Therefore, a Monte Carlo candidate trajectory generation strategy is adopted: learning movement patterns from historical complete migratory bird trajectory data to generate... Article from arrive candidate trajectories ; For the first Candidate trajectories; The total number of candidate trajectories generated; The sign for "much greater than"; the range of "much greater than" is limited to: ,Right now ;

[0113] Calculate a confidence score, representing:

[0114] ;

[0115] In the formula, For the first Candidate trajectories Confidence score; Let m be the velocity of the m-th interpolation point on the candidate trajectory; These are velocity-type motion parameters estimated from migratory bird trajectory data; In order to be in Down, speed Probability distribution of occurrence; Let m be the angular velocity of the m-th interpolation point on the candidate trajectory; Angular velocity-type motion parameters estimated from migratory bird trajectory data; In order to be in Down, The probability distribution of angular velocity occurrence; Let be the vertical velocity of the m-th interpolation point on the candidate trajectory; Vertical velocity-type motion parameters estimated from migratory bird trajectory data; In order to be in Downward, vertical velocity Probability distribution of occurrence;

[0116] ;right arrive Compare and select the candidate trajectory with the highest score. ,Will As the interpolation result of the missing segment, i.e. ; Estimate the position of the interpolation point at time t; This is the interpolation result for the missing segment at time t.

[0117] Step S2: Based on the migratory bird trajectory data after data preprocessing and adaptive interpolation, define the temporal locality index, duration potential, spatial dispersion index, and spatial heterogeneity index; weight and fuse the temporal locality index, duration potential, spatial dispersion index, and spatial heterogeneity index to obtain a comprehensive spatiotemporal rest confidence function; compare the comprehensive spatiotemporal rest confidence function with the set rest judgment threshold to obtain the set of high-confidence rest points for individual migratory birds;

[0118] Each data point in the migratory bird trajectory data after data preprocessing and adaptive interpolation To define the temporal locality index, the duration potential, the spatial dispersion index, and the spatial heterogeneity index; For the first migratory bird trajectory data after data preprocessing and adaptive interpolation There are 10 data points; the data points include interpolation points and original observation points.

[0119] Define a temporal locality index, which represents:

[0120] ;

[0121] In the formula, for Temporal locality indicators; for Inner time dimension t e Standard deviation; for Inner time dimension t e The mean; Here, is the regularization parameter; ; For the first The time window corresponding to each data point; For the first The timestamps corresponding to each data point; The time window is half the width;

[0122] Define duration potential, representing:

[0123] ;

[0124] In the formula, for The potential for duration; This represents the number of neighboring points; For the first Timestamps of each data point; This is the duration decay coefficient;

[0125] Define the spatial scattering index, which represents:

[0126] ;

[0127] In the formula, for Spatial dispersion index; For the same time window The standard deviation of the longitude and latitude coordinates of all points within the plane; Spatial scale parameter;

[0128] Define a spatial heterogeneity index, which represents:

[0129] ;

[0130] In the formula, for Spatial heterogeneity index; In order to be in The number of points within the spatial neighborhood; for with neighbors Spatial distance between them; For the first migratory bird trajectory data after data preprocessing and adaptive interpolation One data point; This is the average distance; It is a very small positive number.

[0131] The specific process for obtaining the comprehensive spatiotemporal pause confidence function is as follows:

[0132] The comprehensive spatiotemporal pause confidence function is obtained, which represents:

[0133] ;

[0134] In the formula, for The comprehensive spatiotemporal pause confidence function; , , , They are respectively , , , Weighting coefficients; To obtain The minimum value between 1 and 1; These are the statistical characteristics of spatial distribution.

[0135] The specific process for obtaining the set of high-confidence resting points for individual migratory birds is as follows:

[0136] Set a global pause threshold ;when At that time, Determined as a rest stop ; ; For the first A rest stop; based on Obtain the set of high-confidence resting points for individual migratory birds. ; This includes the estimated resting points for all decisions.

[0137] Step S3: Perform cluster analysis on the set of high-confidence stopover points of migratory birds to obtain preliminary clusters. Perform saliency filtering on the preliminary clusters and output the set of core habitat clusters.

[0138] The specific process of obtaining the preliminary clustering is as follows:

[0139] right Perform spatial kernel density estimation, spatial location ,express:

[0140] ;

[0141] In the formula, For spatial location The kernel density estimate at that location; for The total estimated number of stopping points; For the first The adaptive bandwidth corresponding to each estimated resting point; for The first in An estimated resting point; For standardized kernel functions;

[0142] Among them, the calculation of the first Adaptive bandwidth ,express:

[0143] ;

[0144] In the formula, The baseline bandwidth; For sensitivity parameters; In order to be in The initial density estimate obtained at the location;

[0145] get after, In spatial location The second-order partial derivatives are used to construct the density peak point matrix;

[0146] Finding through numerical optimization algorithms Density peak points where the gradient is zero, i.e. The density peak points represent the points where the gradient is zero. The density peak point matrix is ​​constructed from the second-order partial derivatives based on the density peak points where the gradient is zero. When the density peak point matrix is ​​negative definite, the density peak points are confirmed as local density maxima. Local density maxima are recorded as candidate cluster centers.

[0147] Will Assigning candidates to cluster centers using a density attraction-based allocation strategy. Iteratively execute the following gradient ascent process to find The convergent density peak point indicates:

[0148] ;

[0149] In the formula, For the first The estimated spatial location of the resting point after the next iteration; For the first The estimated spatial location of the resting point after the next iteration; This is the iteration step size; for exist The gradient vector at that point; Let be the norm of the gradient vector;

[0150] right The iterative process continues until... If the value is less than the set minimum threshold, stop the iteration. If the convergence points reach a single density peak, then the convergent density peak points are assigned to the same cluster. Preliminary clustering was completed; For the first Clusters.

[0151] The specific process of outputting the core habitat cluster set is as follows:

[0152] Saliency filtering is performed on the initial clustering to determine the significance of each cluster. Calculate significance score ,express:

[0153] ;

[0154] In the formula, For clusters The significance score; For the first Cluster The number of density peak points contained therein; For all clusters The average number of density peak points; For clusters The maximum value of the kernel density estimate for all density peak points within the range; The standard deviation of the kernel density estimate; For clusters The minimum value of the kernel density estimate for all density peak points within the range;

[0155] Set a significance threshold ,when At that time, retain cluster;

[0156] The remaining clusters are processed using the α-shape algorithm. Perform calculations using the set parameters. (Value is 0.1), based on the estimated spatial distribution of resting points, a suitable polygon boundary is generated. The polygon boundary determines the habitat core clusters, and the set of habitat core clusters is output. ; For the first A collection of core habitat clusters; For the first A cluster of core habitat areas;

[0157] The core control parameters of the α-shape algorithm are obtained through... Adjust the fit tightness between the polygon boundary and the spatial distribution of the resting points (value is 0.1); Parameter The smaller the value, the closer the polygon boundary is to the preserved cluster. ,parameter The larger the value, the closer the polygon boundary is to the convex hull (a cluster of non-convex shapes).

[0158] Step S4: Based on the output habitat core area cluster set, construct the cluster feature vector, calculate the cluster feature vector using a weighted metric, obtain the inter-cluster dissimilarity matrix, perform spatial clustering fusion on the inter-cluster dissimilarity matrix, and obtain the fused super cluster set, which serves as the flight path of migratory bird trajectory data;

[0159] The specific process for obtaining the inter-cluster dissimilarity matrix is ​​as follows:

[0160] by Using the basic unit, construct the cluster's feature vector, representing:

[0161] ;

[0162] In the formula, For the first The feature vectors of each cluster; For the first The longitude coordinates of each cluster; For the first The weighted center latitude coordinates of each cluster; For the first The area of ​​each cluster; For the first The total number of rest points contained in a cluster; For the first The average stopping confidence of all stopping points within a cluster; For the first The sum of the total dwell times of all resting points within a cluster;

[0163] Measure any two clusters and The dissimilarity and distance between them are calculated using a weighted metric that combines geographic distance and attribute distance, expressed as follows:

[0164] ;

[0165] In the formula, For clusters with cluster The overall dissimilarity between them; This represents the weighting coefficient for geographical distance; For clusters with cluster Geographical distance between center points; This is the normalization factor for geographical distance; This is the weighting coefficient for attribute distance; for and The distance between the attributes; For the first A collection of core habitat clusters; For the first The feature vectors of each cluster;

[0166] Will Arranged in row-column form, they form a cluster dissimilarity matrix.

[0167] The specific process for obtaining the fused supercluster set is as follows:

[0168] The Average Linkage method from Hierarchical Clustering is used to analyze the clusters in the inter-cluster dissimilarity matrix. with cluster Perform cluster fusion to obtain a fused cluster set;

[0169] Set fusion distance threshold To merge all the clusters into one cluster with cluster The distance between them is less than The clusters are merged to obtain a supercluster set. ; This is the e-th supercluster after merging;

[0170] For each supercluster in the merged supercluster set First find the matching Another recent supercluster ,calculate and Distance between the centroids ;

[0171] Set the maximum isolation distance threshold ,when At that time; then it is believed Too isolated in space This likely corresponds to occasional emergency stops (such as temporary shelter during severe weather), data noise, or the unique behavioral patterns of a very small number of migratory birds. Remove; when At that time, the merged supercluster set Flight paths used as data for migratory bird trajectories.

[0172] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying migratory bird tracking habitat based on spatio-temporal fusion and adaptive kernel density, characterized in that, The method comprises the following steps: Step S1: uniform processing is performed on the sampled migratory bird trajectory data, and a three-level interpolation strategy is used for interpolation reconstruction on the uniform processed migratory bird trajectory data to obtain migratory bird trajectory data after data preprocessing and adaptive interpolation; the three-level interpolation strategy for interpolation reconstruction is specifically: when it is detected that the migratory bird trajectory data is missing, cubic Hermite spline is used for interpolation reconstruction; when the migratory bird trajectory data is complex, neighborhood weighted interpolation is used; when the missing section of the migratory bird trajectory data is extremely long and reliable reference data is lacking, Monte Carlo candidate trajectory is used for interpolation reconstruction; Step S2: based on the migratory bird trajectory data after data preprocessing and adaptive interpolation, a time locality index is defined, a duration potential is defined, a spatial dispersion index is defined, and a spatial heterogeneity index is defined; the time locality index, the duration potential, the spatial dispersion index, and the spatial heterogeneity index are weighted and fused to obtain a comprehensive spatiotemporal stop confidence function; the comprehensive spatiotemporal stop confidence function is compared with a set stop determination threshold to obtain a high-confidence stop point set of the migratory bird individual; Step S3: the high-confidence stop point set of the migratory bird individual is subjected to cluster analysis processing to obtain preliminary clustering, and the preliminary clustering is subjected to significance filtering to output a habitat core area cluster set; Step S4: based on the output habitat core area cluster set, a feature vector of the cluster is constructed, the feature vector of the cluster is subjected to weighted measurement calculation to obtain a inter-cluster dissimilarity matrix, and the inter-cluster dissimilarity matrix is subjected to spatial cluster fusion to obtain a fused super cluster set as a flight path of the migratory bird trajectory data; The specific process of obtaining the fused super cluster set is as follows: The average linkage method in the hierarchical clustering algorithm is used to cluster and fuse the cluster differences in the cluster and the cluster to obtain a fused cluster set; is the first habitat core cluster set; is the first habitat core cluster set; Setting fusion distance threshold , all the clusters in the set of fused clusters , the clusters whose mutual distance between clusters is less than are merged to obtain the set of fused super clusters is the e-th fused super cluster​ For each supercluster in the merged supercluster set First find the matching Another recent supercluster ,calculate and Distance between the centroids ; Setting a maximum isolated distance threshold When , the culling; when , the fused super cluster set is taken as the flight path of the candidate bird trajectory data.

2. The method of claim 1, wherein the method is based on spatio-temporal fusion and adaptive kernel density for migratory bird tracking habitat identification. The specific process of step S1 is as follows: The sampled migratory bird trajectory data is imported into a unified data framework, and the migratory bird trajectory data is represented as: ； wherein, is the timestamp of the i-th migratory bird at the k-th trajectory point; is the horizontal coordinate of the i-th migratory bird at the k-th trajectory point; is the vertical coordinate of the i-th migratory bird at the k-th trajectory point; is the height of the i-th migratory bird at the k-th trajectory point; is the positioning quality indicator of the i-th migratory bird at the k-th trajectory point; is the total number of trajectory points of the i-th migratory bird; uniforming the irregularly sampled migrant track data to a fixed time resolution, i.e. a target step size ; The position estimation of the migratory bird at the interpolation point t is calculated, and the time window width is adaptively determined according to the sampling interval distribution of the migratory bird trajectory data , and the time window is ; there are original observation points in the time window, the time stamp of the original observation point is , and the corresponding position is ; , the position estimation of the migratory bird at the interpolation point t is calculated based on , and the position estimation of the migratory bird at the interpolation point t is calculated based on , which indicates that ； wherein is the position estimate of the migratory bird at time t for the interpolated point; is the weight of the th original observation point; is the weight of the th original observation point at time stamp corresponding actual position; wherein represents: ； wherein is a very small positive number; is the absolute value of the time difference; The specific process of the three-level interpolation strategy is as follows: Let the two endpoints of the interpolation interval for the migratory bird's trajectory be respectively... and When the interval between the two endpoints of the interpolation interval satisfy This indicates that migratory bird trajectory data is missing; The timestamp of the left endpoint of the interpolation interval; for The actual location of migratory birds at all times; The timestamp of the right endpoint of the interpolation interval; for The actual location of migratory birds at all times; The time interval between the two endpoints of the interpolation interval; When the missing of the migratory bird trajectory data is indicated, a cubic Hermite spline with end-point speed constraint is used to interpolate and reconstruct, Constructing representation: ； wherein is the left end position basis function; is the right end position basis function; is the left end velocity basis function; is the right end velocity basis function; is the interpolated left end position of the instantaneous velocity vector of the migratory bird at time t; is the interpolated right end position of the instantaneous velocity vector of the migratory bird at time t; is the normalized time parameter; The method of sampling the cubic Hermite spline guarantees that the and satisfies the dynamics of the short-time motion of the migratory bird; is the position estimate of the migratory bird at time is the position estimate of the migratory bird at time is the position estimate of the migratory bird at time is the position estimate of the migratory bird at time is the instantaneous velocity vector of the migratory bird at time is the instantaneous velocity vector of the migratory bird at time is the instantaneous velocity vector of the migratory bird at time is the instantaneous velocity vector of the migratory bird at time When neighborhood weighted interpolation is adopted to introduce behavior similarity adjustment on the basis of , the result is , which means: ； wherein, is the interpolation weight of the jth original observation point after introducing the behavior similarity adjustment; is the weight of the jth original observation point; is the natural exponential function; is the difference between the jth original observation point and the jth historical observation point; is the difference between the jth original observation point and the jth historical observation point; is the difference between the jth original observation point and the jth historical observation point; is the motion feature vector of the jth historical observation point in the local window; is the motion feature vector of the interpolation point t in the local window; is the variance parameter; When , it indicates that the missing section of the migratory trajectory data is extremely long and lacks reliable reference, and the Monte Carlo candidate trajectory generation strategy is adopted: learn the motion pattern from the historical complete migratory trajectory data to generate candidate trajectories from to ; is the first candidate trajectory; is the total number of generated candidate trajectories; is a symbol much larger than , that is, ; A confidence score is calculated, which is represented as: ； wherein is the confidence score of the candidate trajectory; is the velocity of the mth interpolation point on the candidate trajectory; is a velocity-like motion parameter estimated from the candidate trajectory data; is the probability distribution of the velocity occurring under is the probability distribution of the angular velocity occurring under is the angular velocity of the mth interpolation point on the candidate trajectory; is an angular velocity-like motion parameter estimated from the candidate trajectory data; is the probability distribution of the angular velocity occurring under is the vertical velocity of the mth interpolation point on the candidate trajectory; is a vertical velocity-like motion parameter estimated from the candidate trajectory data; is the probability distribution of the vertical velocity occurring under is the probability distribution of the vertical velocity occurring under ; compare to select the candidate trajectory with the highest score , the interpolation result of the missing segment, i.e. ; ; the position estimate at the interpolation point t; the interpolation result of the missing segment at t.

3. The method of claim 2, wherein the method is based on spatio-temporal fusion and adaptive kernel density for migratory bird tracking habitat identification. The specific process of defining the time locality index, the duration potential, the spatial dispersion index, and the spatial heterogeneity index is as follows: each data point in the pre-processed and interpolated migratory bird trajectory data to define a temporal locality indicator, to define a duration potential, to define a spatial dispersion index, to define a spatial heterogeneity indicator; is a data point in the pre-processed and interpolated migratory bird trajectory data ; the data points include interpolated points and original observed points; The time locality index is defined, which is represented as: ； In the formula, for Temporal locality indicators; for Inner Time Dimension Standard deviation; for Inner Time Dimension The mean; Here, is the regularization parameter; ; For the first The time window corresponding to each data point; For the first The timestamps corresponding to each data point; The time window is half the width; The duration potential is defined, which is represented as: ； wherein, is the duration potential; is the number of adjacent points; is the timestamp of the th data point; is the duration decay coefficient; The spatial dispersion index is defined, which is represented as: ； wherein is the spatial dispersion index; is the standard deviation of the longitude and latitude coordinates of the planar positions of all points within the same time window is the spatial scale parameter;​​ The spatial heterogeneity index is defined, which is represented as: ； In the formula, for Spatial heterogeneity index; In order to be in The number of points within the spatial neighborhood; for with neighbors Spatial distance between them; For the first migratory bird trajectory data after data preprocessing and adaptive interpolation One data point; This is the average distance; It is a very small positive number.

4. The method of claim 3, wherein: The specific process of obtaining the comprehensive spatiotemporal stop confidence function is as follows: The comprehensive spatiotemporal stop confidence function is obtained, which is represented as: ； wherein is a combined spatio-temporal dwell confidence function; , , , are , , , weighting coefficients; is the minimum between the value and the value 1; is a statistical feature of the spatial distribution.

5. The method of claim 4, wherein: The specific process of obtaining the high-confidence stop point set of the migratory bird individual is as follows: Setting a global rest criterion threshold ; when , a rest point is determined ; ; ; is the first rest point; based on , a high confidence rest point set for the individual is obtained ; ; includes all determined estimated rest points.

6. The method of claim 5, wherein: The specific process of obtaining the preliminary clustering is as follows: To perform a spatial kernel density estimation, spatial locations , are represented: ； wherein is a kernel density estimate for spatial locations ; is a total number of estimated stall points in ; is an adaptive bandwidth corresponding to the th estimated stall point; is the th estimated stall point in ; is a normalized kernel function; obtained after, at spatial positions where the second-order partial derivatives constitute a density peak point matrix; Finding through numerical optimization algorithms Density peak points where the gradient is zero, i.e. The density peak points represent the points where the gradient is zero. The density peak point matrix is ​​constructed from the second-order partial derivatives based on the density peak points where the gradient is zero. When the density peak point matrix is ​​negative definite, the density peak points are confirmed as local density maxima. Local density maxima are recorded as candidate cluster centers. Will Assigning candidates to cluster centers using a density attraction-based allocation strategy. Iteratively execute the following gradient ascent process to find The convergent density peak point indicates: ； In the formula, For the first The estimated spatial location of the resting point after the next iteration; For the first The estimated spatial location of the resting point after the next iteration; This is the iteration step size; for exist The gradient vector at that point; Let be the norm of the gradient vector; to The iterative process continues until less than a set minimum threshold, the iteration is stopped, converges to a density peak point, the same density peak point to which the convergence is assigned to belong to the same cluster , the preliminary clustering is completed; for the first cluster.

7. The method of claim 6, wherein: The specific process of outputting the habitat core area cluster set is as follows: Significance filtering of the preliminary clustering, for each cluster Computing the significance score denotes: ； wherein is the distinctness score of the cluster ; is the number of density peak points contained in the th cluster ; is the average of the number of density peak points of all clusters ; is the maximum of the kernel density estimate of all density peak points within the cluster ; is the standard deviation of the kernel density estimate; is the minimum of the kernel density estimate of all density peak points within the cluster ; Setting a significance threshold When the cluster is retained clusters; The retained clusters are processed using the α-shape algorithm. Perform calculations using the set parameters. Based on the estimated spatial distribution of resting points, a suitable polygonal boundary is generated. The polygonal boundary determines the core habitat clusters, and the set of core habitat clusters is output. ; For the first A cluster of core habitat areas, The core control parameters of the α-shape algorithm are obtained through... Adjust the tightness of the spatial distribution of polygon boundaries and rest points.

8. The method of claim 7, wherein: The specific process of obtaining the inter-cluster dissimilarity matrix is as follows: With as a basic unit, the feature vector of the cluster is constructed, which is represented as: ； wherein is the feature vector of the th cluster; is the longitude coordinate of the th cluster; is the weighted center latitude coordinate of the th cluster; is the area of the th cluster; is the total number of stop points contained in the th cluster; is the average stop confidence of all stop points in the th cluster; is the sum of total stay duration of all stop points in the th cluster; Measuring the dissimilarity and distance between any two clusters and is represented by a weighted metric computation that combines geographical distance and attribute distance. ； wherein, is the integrated dissimilarity between the cluster and the cluster ; is a weight coefficient for the geographical distance; is the geographical distance between the cluster and the cluster center point; is a normalization factor for the geographical distance; is a weight coefficient for the attribute distance; is the attribute distance between the cluster and the cluster ; is the feature vector of the th cluster; The The row-column arrangement forms a dissimilarity matrix between clusters.

Images