Animal population quantity prediction method and system and storage medium

Abstract

The invention discloses an animal population quantity prediction method and system and a storage medium, and the method comprises the steps: obtaining a target animal population habitat remote sensing image, and carrying out the pixel-by-pixel feature extraction, and obtaining a plurality of pixel feature matrixes; matrix feature selection is carried out based on a sparse feature selection method, and a simplified matrix is constructed; selecting a plurality of clustering center points from the simplified matrix; according to the selection condition of the plurality of clustering center points, the feature information on the plurality of data points and a preset classification model, feature category clustering is carried out based on an alternate optimization mode, and a plurality of area images and corresponding categories thereof are obtained; performing calculation according to each regional image and the corresponding category thereof and a preset density model to obtain a population density corresponding to each regional image; and performing analysis according to the category and population density corresponding to each region image, and determining an animal population quantity prediction result of the target animal population. The embodiment of the invention can improve the accuracy of animal population quantity prediction, and can be widely applied to the technical field of image recognition.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, system and storage medium for predicting animal population size. Background Technology

[0002] Population prediction is the process of forecasting the population size of one or more species over a specific time period using mathematical models and data analysis methods. This prediction typically relies on historical data, environmental factors, reproductive rates, mortality rates, migration patterns, and other factors to estimate future population trends. Population prediction is widely used in ecology, environmental protection, and species management to help policymakers understand the dynamic changes in ecosystems, formulate scientific species conservation measures, and avoid problems such as ecological imbalance, species extinction, or biological invasion. Through scientific population prediction, natural resources can be better managed, biodiversity can be protected, ecological balance can be maintained, and the efficiency of ecological restoration can be improved.

[0003] With the development of remote sensing technology, ecological information extraction methods based on high-resolution remote sensing imagery are increasingly being applied to population size prediction. While remote sensing imagery offers advantages such as wide spatial coverage, rapid update cycles, and non-contact data acquisition, existing research has limited accuracy in identifying complex remote sensing images, leading to inaccurate population size predictions.

[0004] Therefore, the accuracy of current technologies for predicting animal population size needs to be improved. Summary of the Invention

[0005] In view of this, in order to solve one of the above problems, the purpose of the embodiments of the present invention is to provide an animal population size prediction method, system and storage medium, which can effectively improve the accuracy of animal population size prediction.

[0006] On one hand, embodiments of the present invention provide a method for predicting animal population size, including:

[0007] Acquire remote sensing images of the animal habitat of the target animal population in the target area; extract features from the feature information of each pixel in the animal habitat remote sensing image to obtain a feature matrix corresponding to each pixel;

[0008] Based on the sparse feature selection method, feature selection is performed on the feature matrix corresponding to each pixel, and a simplified matrix containing several data points is constructed according to the feature selection result; the feature information on the several data points is determined by the feature selection result of the feature matrix corresponding to the several pixels.

[0009] Several cluster centers are selected from the simplified matrix; based on the selection of several cluster centers, the feature information on several data points, and the preset classification model, feature category clustering is performed based on an alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image;

[0010] The population density corresponding to each habitat area image is calculated based on several habitat area images and a preset animal density model.

[0011] The population size prediction results of the target animal species are determined by analyzing the habitat species and population density corresponding to several habitat area images.

[0012] Specifically, the feature information of the pixel includes its spectral reflectance values ​​in several spectral bands; the simplified matrix is ​​constructed in the following manner:

[0013] Based on the spectral reflectance values ​​of each pixel in several bands, the band signal-to-noise ratio of each pixel in each band is calculated, and the auxiliary weight vector corresponding to each pixel is determined according to the calculation results.

[0014] Based on a preset sparse feature matrix and an auxiliary weight vector corresponding to each pixel, feature selection is performed on the feature matrices of several pixels to obtain a simplified feature matrix corresponding to each pixel.

[0015] Construct a simplified matrix containing several data points; the feature information on the several data points corresponds one-to-one with the simplified feature matrix corresponding to the several pixels.

[0016] Specifically, the expression for the preset sparse feature matrix is ​​as follows:

[0017]

[0018] Among them, s T (diag(M)+αw) is the evaluation function of the feature matrix corresponding to the pixel; diag(M) is the importance measure of each feature in the feature matrix corresponding to the pixel; α is the weight coefficient, w is the auxiliary weight vector; d represents the number of features selected from the feature matrix corresponding to the pixel.

[0019] Specifically, the feature information on the data points includes the feature information on the cluster centers; the step of clustering feature categories based on the selection of the cluster centers, the feature information on the data points, and a preset classification model, using an alternating optimization method, to obtain several habitat area images and the habitat type information corresponding to each habitat area image, includes:

[0020] Based on a preset loss function, the selection of several cluster centers, the feature information of several data points, the feature information of several cluster centers, and a preset classification model, feature category clustering is performed based on an alternating optimization method to determine the habitat type corresponding to each data point.

[0021] The pixels corresponding to data points of the same habitat species are merged according to their spatial location to obtain several habitat area images and their corresponding habitat species.

[0022] Specifically, the habitat type corresponding to each pixel is determined in the following way:

[0023] Based on the feature information of the data points and the feature information of each cluster center point, the distance between the data points and each cluster center point is calculated; the distance calculation method includes Euclidean distance algorithm and cosine similarity algorithm.

[0024] Set the initial classification level weight of the data points; calculate the similarity between the data points and each cluster center point based on the preset loss function, the initial classification level weight of the data points, and the distance between the data points and each cluster center point;

[0025] Based on the similarity between the data points and each cluster center, the data points are clustered by feature category to determine the feature category clustering result of the data points;

[0026] Adjust the selection of several cluster centers; re-cluster each data point according to a preset formula and adjust the classification weight of each data point according to the clustering results of each data point until the clustering results of each data point tend to be stable; determine the habitat type corresponding to the data point according to the adjusted clustering results of the data points.

[0027] Specifically, the expression for the preset loss function is:

[0028]

[0029] Where tensor A represents the reconstruction error matrix, determined by the distance between the i-th data point and each cluster center point, ||A|| σ γL represents the weighted norm result of tensor A under the robust adjustment factor σ; i L represents the classification rank weight of the i-th data point, γ is the global scaling factor, and L i It is a local rank factor; α iThe eigenvector of the i-th data point is determined by the feature information of the data point; ||α i ||2 is the vector magnitude of the feature vector of the i-th data point; Let L2 be the squared L2 norm of the eigenvector of the i-th data point.

[0030] Specifically, the expression of the preset formula is:

[0031] max W,F,G Tr(W T S t W)-α||Γ 1 / 2 (X T W-FG T )|| σ ;

[0032] in, Let Tr(W) be the optimization function for vectors W, F, and G, where W represents the feature matrix of the data points, F represents the clustering result of the feature categories of the data points, and G represents the selection of several cluster centers; T S t W) is the inter-class divergence target, determined by the degree of distribution difference in the clustering results of the corresponding feature categories for each data point; S t The covariance matrix measures the degree of difference in the distribution of clustering results for the corresponding feature categories of each data point; (X T W-FG T ) represents the reconstruction error, determined by the distance between data points and each cluster center; α is the control weight, used to balance the inter-class divergence objective and the reconstruction error; Γ 1 / 2 This is a weight matrix used to represent the importance of the data points in the loss function.

[0033] Specifically, the model calculation parameters of the preset animal density model are confirmed in the following way:

[0034] Acquire several habitat area image samples; perform adjustment processing on each of the habitat area image samples to obtain several computational image samples; the adjustment processing includes data cleaning, missing value handling, and normalization processing;

[0035] Based on each computational image sample, determine the population size information and habitat type information corresponding to each computational image sample;

[0036] Based on each computational image sample and the population size information of each computational image sample, pixel-by-pixel calculations are performed to determine the population density information corresponding to each pixel unit of each computational image sample.

[0037] Regression calculations are performed based on the habitat type information corresponding to each calculated image sample and the population density information corresponding to each pixel unit of each calculated image sample. The model calculation parameters of the preset animal density model are determined based on the regression calculation results.

[0038] On the other hand, embodiments of the present invention also provide an animal population size prediction system, comprising:

[0039] The feature extraction module is used to acquire remote sensing images of the animal habitat of the target animal population in the target area; and to extract features from the feature information of each pixel in the remote sensing image of the animal habitat to obtain a feature matrix corresponding to each pixel.

[0040] The feature selection module is used to perform feature selection on the feature matrix corresponding to each pixel based on the sparse feature selection method, and construct a simplified matrix containing several data points based on the feature selection result; the feature information on the several data points is determined by the feature selection result of the feature matrix corresponding to the several pixels;

[0041] The classification prediction module is used to select several cluster centers from the simplified matrix; based on the selection of several cluster centers, the feature information on several data points and the preset classification model, feature category clustering is performed based on an alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image;

[0042] Density prediction module: Calculates the population density corresponding to each habitat area image based on each habitat area image, its corresponding habitat species, and a preset animal density model;

[0043] The analysis module is used to analyze the habitat species and population density corresponding to several habitat area images to determine the predicted animal population size of the target animal population.

[0044] On the other hand, embodiments of the present invention also provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the above-described method.

[0045] Implementing the embodiments of the present invention has the following beneficial effects:

[0046] This invention provides a method, system, and storage medium for predicting animal population sizes. The method acquires remote sensing images of the target animal population and extracts features pixel-by-pixel from these images to obtain a feature matrix for each pixel. A sparse feature selection method is then used to select features from the feature matrix for each pixel, thereby constructing a simplified matrix containing several data points. Next, several cluster centers are selected from the simplified matrix, and based on the selection of cluster centers, the feature information of the data points, and a preset classification model, feature category clustering is performed using an alternating optimization method to obtain several habitat area images and the habitat species corresponding to each habitat area image. Further, based on each habitat area image, its corresponding habitat species, and a preset animal density model, the population density corresponding to each habitat area image is calculated. Furthermore, by combining the habitat species and population density information corresponding to several habitat area images, the population size prediction results of the target animal population are determined. To this end, on the one hand, this invention uses a sparse feature selection method to retain features in complex images that are more helpful for classification, improving the accuracy of subsequent habitat species prediction; on the other hand, this invention uses an alternating optimization method for feature category clustering, optimizing the clustering effect by adjusting the selection of cluster centers, thus improving the accuracy of habitat species prediction using complex images. Finally, this invention combines the habitat species and population density information of the target animal population, analyzing both together to more accurately determine changes in the target animal population's habitat location and population density, effectively improving the accuracy of animal population size prediction. Attached Figure Description

[0047] Figure 1 This is a schematic flowchart of the steps of an animal population size prediction method provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of a population habitat area distribution image provided in an embodiment of the present invention;

[0049] Figure 3 This is a structural block diagram of an animal population prediction system provided in an embodiment of the present invention;

[0050] Figure 4 This is a structural block diagram of an animal population prediction device provided in an embodiment of the present invention. Detailed Implementation

[0051] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0052] The following is an explanation of several terms used in this application:

[0053] Image band range: This refers to the range of spectral information represented by each pixel in an image. In remote sensing and multispectral image processing, images are typically composed of multiple bands, each representing a different spectral range, such as visible light, near-infrared, and short-wave infrared. Each band captures electromagnetic wave information of different wavelengths, and the band range determines the spectral area that band can sense. Combinations of different bands can provide different information about the surface features of objects or ground features, helping to conduct more accurate image analysis.

[0054] Spectral efficiency refers to the ability of an object or material to reflect, transmit, or emit light within different wavelength ranges. It is usually expressed in terms of wavelength, describing the intensity of an object's response in various spectral bands (such as visible light, infrared, ultraviolet, etc.).

[0055] Spectral reflectance refers to the ability of an object's surface to reflect light within a specific wavelength range, usually expressed as the ratio of reflected light intensity to incident light intensity. It reflects the object's reflectivity to different wavelengths of light and is commonly used in remote sensing and land cover classification. Different types of land cover (such as water bodies, vegetation, and soil) have different spectral reflectance values ​​in different spectral bands; these differences can be used to distinguish land cover types or assess environmental changes. Spectral reflectance is a key parameter in remote sensing image analysis, helping to extract and identify target information.

[0056] Band Signal-to-Noise Ratio (SNR): Band signal-to-noise ratio (SNR) refers to the ratio (usually expressed in decibels, dB) of the average power of a signal to the average power of noise within a specific frequency range. This metric quantifies the strength of a signal relative to background noise within a target frequency band and directly affects the bit error rate of a communication system, the clarity of remote sensing images, and the accuracy of radar detection. A high band SNR is a key parameter for ensuring data transmission reliability and improving imaging quality, and it has core application value in fields such as wireless communication, satellite remote sensing, and acoustic detection.

[0057] Alternating optimization: a common technique for solving complex optimization problems involving multiple subsets of variables. Its core idea is that when it is difficult to directly and jointly optimize all variables, the other subsets are fixed, and optimization is performed only on the currently selected subset. Then, the optimized subset is switched alternately, and this process is repeated. By alternately and iteratively updating different variable blocks, the optimal solution to the problem is gradually approximated.

[0058] Sparse feature selection methods: In remote sensing image processing, this method aims to automatically select the most discriminative feature subset from high-dimensional spectral, spatial, or temporal features while suppressing redundant or noisy information. By introducing a sparse feature selection matrix or sparse constraints (such as L1 regularization, group sparsity, or low-rank constraints), this method forces the weights of irrelevant features to zero in tasks such as classification, target detection, or land cover recognition, thereby reducing computational burden, enhancing model generalization ability, and improving interpretability. Its advantage lies in its ability to adaptively mine joint spatial and spectral features, highlighting key bands or spatial regions, and is suitable for dimensionality reduction and feature optimization of high-resolution, multispectral / hyperspectral remote sensing data.

[0059] Quantitative regression models are statistical models used to predict continuous variables. They aim to estimate the value of a dependent variable using a set of independent variables (predictors). Common quantitative regression models include linear regression, ridge regression, and LASSO regression. In linear regression, the model predicts the value of the dependent variable by fitting a linear relationship to the data. Quantitative regression is widely used in economics, marketing, engineering, and biology to analyze causal relationships and predict future trends.

[0060] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for predicting animal population size, which includes the following steps S100 to S500.

[0061] S100: Acquire remote sensing images of the animal habitat of the target animal population in the target area; extract features from the feature information of each pixel in the remote sensing image of the animal habitat to obtain the feature matrix corresponding to each pixel.

[0062] For remote sensing images of animal habitats of target animal populations in the target area, pixel-by-pixel feature extraction is performed to extract feature information for each pixel, resulting in a feature matrix for each pixel, which serves as the input sample for clustering. The pixel-level feature information includes multidimensional features such as band ratios, color space features, and texture features.

[0063] S200: Based on the sparse feature selection method, feature selection is performed on the feature matrix corresponding to each pixel, and a simplified matrix containing several data points is constructed according to the feature selection result; the feature information on several data points is determined by the feature selection result of the feature matrix corresponding to several pixels.

[0064] Based on the sparse feature selection method, feature selection is performed on the feature matrix of each pixel in the remote sensing image to construct the corresponding simplified matrix. The resulting simplified matrix can effectively retain the most useful features for each pixel, thereby reducing the impact of irrelevant image features in complex images on subsequent model predictions or calculations.

[0065] S300: Select several cluster centers from the simplified matrix; based on the selection of several cluster centers, the feature information on several data points, and the preset classification model, perform feature category clustering based on the alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image.

[0066] We set an initial set of cluster centers (initial representative points), assuming predefined habitat classes to be divided into (e.g., snow, rock, water, etc.). Each class has a "reference center point" representing the typical characteristics of that class. Using an alternating optimization approach, we adjust the influencing factors in feature class clustering (e.g., the selection of cluster centers) to stabilize the clustering results and ensure the overall classification performance meets predefined requirements. This yields several habitat region images and the corresponding habitat types for each image. For example... Figure 2 The image shown is a population habitat distribution image provided by the present invention, displaying several habitat area images and a summary of the habitat types corresponding to each habitat area image.

[0067] S400: Calculate the population density corresponding to each habitat area image based on the image of each habitat area, the corresponding habitat species, and the preset animal density model.

[0068] By combining the habitat species information corresponding to the habitat area images with the preset animal density model to calculate the population density of each habitat area image, the accuracy of population density prediction for the target animal population can be effectively improved.

[0069] S500: Based on the analysis of habitat species and population density corresponding to several habitat area images, determine the predicted animal population size of the target animal species.

[0070] The system summarizes the population data for each habitat area, determines the distribution of populations among different species in each habitat, and outputs the predicted population size of the target animal species.

[0071] Specifically, in step S100, before feature extraction, the remote sensing image can be preprocessed; the methods of image preprocessing include, but are not limited to, filtering, normalization and denoising.

[0072] In some embodiments, in step S200, the feature information on a pixel includes its spectral reflectance values ​​in several bands; the simplified matrix is ​​constructed in the following manner:

[0073] S210: Based on the spectral reflectance values ​​of each pixel in several bands, calculate the band signal-to-noise ratio of each pixel in each band, and determine the auxiliary weight vector corresponding to each pixel based on the calculation results;

[0074] Since remote sensing images have many spectral ranges, not every spectral range is helpful in distinguishing habitat types. Therefore, in this embodiment of the invention, by calculating the signal-to-noise ratio of a pixel in each spectral range, the most discriminative spectral range is selected from all spectral ranges of the pixel as a reference, and an auxiliary weight vector corresponding to the pixel is constructed as an auxiliary reference factor for subsequent feature selection.

[0075] S220: Based on the preset sparse feature matrix and the auxiliary weight vector corresponding to each pixel, feature selection is performed on the feature matrix of several pixels to obtain the simplified feature matrix corresponding to each pixel;

[0076] Based on the auxiliary weight vector corresponding to each pixel and the preset sparse feature matrix, the feature matrix of each pixel is filtered to retain a number of effective features and reduce the influence of features that are not important for prediction. Each pixel in the remote sensing image is mapped to a low-dimensional subspace through the sparse feature matrix, forming a more compact and discriminative feature representation.

[0077] S230: Construct a simplified matrix containing several data points; the feature information on several data points corresponds one-to-one with the simplified feature matrix corresponding to several pixels.

[0078] A simplified matrix is ​​constructed as the data basis for subsequent model prediction or calculation. The information on the matrix is ​​determined by the feature information on the simplified feature matrix corresponding to each pixel.

[0079] Specifically, the expression for the preset sparse feature matrix in step S220 is as follows:

[0080]

[0081] Among them, s T (diag(M)+αw) is the evaluation function of the feature matrix corresponding to the pixel; diag(M) is the importance measure of each feature in the feature matrix corresponding to the pixel; α is the weight coefficient, w is the auxiliary weight vector; d represents the number of features selected from the feature matrix corresponding to the pixel.

[0082] In some embodiments, matrix M is an importance assessment result for each band and image feature dimension of pixels in a remote sensing image of an animal habitat, used to indicate which features are more helpful in distinguishing different types of habitat areas. Further, in this embodiment, diag(M) is a vector of diagonal elements of matrix M, used to measure the independent discriminative power of each remote sensing feature (band reflectance, texture value, etc.) in identifying animal habitat categories. This "importance" is reflected in the fact that a certain feature can provide more distinguishing information when differentiating different types of habitat areas (such as bare rock, snow, and dung).

[0083] Specifically, in step S300, the feature information on several data points includes the feature information on several cluster center points; based on the selection of several cluster center points, the feature information on several data points, and a preset classification model, feature category clustering is performed using an alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image, including:

[0084] S310: Based on a preset loss function, the selection of several cluster centers, the feature information of several data points and the feature information of several cluster centers, and a preset classification model, feature category clustering is performed based on an alternating optimization method to determine the habitat type corresponding to each data point.

[0085] The feature information at each data point and the feature information at each cluster center point are used to perform calculations, and the feature category clustering process is optimized according to the alternating optimization method to determine the habitat type corresponding to each data point.

[0086] S320: Merge the pixels corresponding to data points of the same habitat species according to their spatial location to obtain several habitat area images and their corresponding habitat species.

[0087] Data points for the same habitat species are aggregated and merged according to spatial location to form complete habitat area patches (such as a contiguous snowfield or bare rock area). Then, based on cluster labels and prior knowledge, these areas are assigned specific ecological categories, providing a basis for subsequent habitat distribution identification and ecological assessment.

[0088] Specifically, in step S310, the habitat type corresponding to each pixel is determined in the following way:

[0089] S311: Based on the feature information of the data points and the feature information of each cluster center, the distance between the data points and each cluster center is calculated; the distance calculation methods include Euclidean distance algorithm and cosine similarity algorithm.

[0090] The distance between each data point and each cluster center is calculated using the feature information of each data point and the feature information of each cluster center.

[0091] S312: Set the initial classification level weights of the data points; calculate the similarity between the data points and each cluster center based on the preset loss function, the initial classification level weights of the data points, and the distance between the data points and each cluster center.

[0092] To enhance the flexibility of judgment, this embodiment of the invention sets classification level weights for data points. The weight of a data point in the classification can be adjusted according to the distance between the data point and the cluster center, thus weakening the influence of abnormal pixels. In addition, based on the distance between the data point and the cluster center, a preset loss function is used to measure the similarity between each data point and the cluster center, and the influence of some interferences in the image (such as shadows and reflections) on the classification results can be automatically reduced.

[0093] S313: Based on the similarity between the data points and each cluster center, perform feature category clustering on the data points to determine the feature category clustering results of the data points.

[0094] Based on the similarity between a data point and each cluster center, it is assigned to the nearest cluster center, which is the habitat species to which it is most likely to belong, thus obtaining the feature category clustering result of the data point.

[0095] S314: Adjust the selection of several cluster centers; re-cluster each data point according to the preset formula and adjust the classification level weight of each data point according to the clustering results of each data point until the clustering results of each data point tend to be stable, and determine the habitat type corresponding to the data point according to the clustering results of the adjusted data points.

[0096] To improve classification accuracy, we employed an "alternating optimization" approach. Each step first readjusts the classification weights and cluster center positions of the data points based on the current classification results, and then updates the classification label for each data point. This process is repeated until the feature category clustering results for each data point are relatively stable and the overall classification effect is optimal. The adjusted feature category clustering results at this point represent the habitat type corresponding to each data point.

[0097] Specifically, the expression for the preset loss function mentioned in step S312 is:

[0098]

[0099] Where tensor A represents the reconstruction error matrix, which is determined by the distance between the i-th data point and each cluster center point, ||A|| σ γL represents the weighted norm result of tensor A under the robust adjustment factor σ; i L represents the classification rank weight of the i-th data point, γ is the global scaling factor, and L i It is a local ranking factor; α i The eigenvector of the i-th data point is determined by the feature information of the data point; ||α i ||2 is the vector magnitude of the feature vector of the i-th data point; Let L2 be the squared L2 norm of the eigenvector of the i-th data point.

[0100] In some embodiments, tensor A represents the reconstruction error residual matrix generated during the clustering or classification process of remote sensing images, determined by the distance between data points and each cluster center. The reconstruction error residual matrix refers to the difference between the projection result of the data point feature matrix in the low-dimensional subspace and the reconstruction result of the cluster centers. Each row corresponds to the feature error of a data point, and the overall model reflects the degree of fit of the system to the current feature representation and clustering.

[0101] Furthermore, if a pixel is severely affected by interference (such as being covered by shadows, reflections, or snow), its error may become unreliable, and we will adjust its L value accordingly. i We set it to a smaller value to account for less error; while for clear, normal pixels, we give it a higher L value. i This allows it to play a greater role in optimization. γL i The significance lies in dynamically adjusting the influence of each pixel, allowing the model to focus more on the features of reliable areas rather than being overly disturbed by outliers, thereby improving the stability and accuracy of remote sensing image processing of animal habitats.

[0102] Specifically, ||α i ||2 represents the vector magnitude of the feature vector of the i-th data point, used to normalize the residual term. The normalized residual term is calculated by dividing the error value by the sample magnitude when calculating the image error (residual). It is used to eliminate the influence caused by differences in feature scale between different samples.

[0103] Specifically, the expression of the preset formula in step S314 is as follows:

[0104] max W,F,G Tr(W T S t W)-α||Γ 1 / 2 (X T W-FG T )||σ (2)

[0105] in, Let Tr(W) be the optimization function for vectors W, F, and G, where W represents the feature matrix of the data points, F represents the clustering result of the feature categories of the data points, and G represents the selection of several cluster centers; T S t W) is the inter-class divergence target, determined by the degree of distribution difference in the clustering results of the corresponding feature categories for each data point; S t The covariance matrix measures the degree of difference in the distribution of clustering results for the corresponding feature categories of each data point; (X T W-FG T ) represents the reconstruction error, determined by the distance between data points and each cluster center; α is the control weight, used to balance the inter-class divergence objective and the reconstruction error; Γ 1 / 2 This is the weight matrix, used to represent the importance of data points in the loss function.

[0106] Furthermore, in animal habitat identification tasks, pixel classification is based on its location in the feature space. This is achieved through Tr(W T S t The inter-class divergence objective of W) is to maximize inter-class divergence, aiming to enhance the differences between different habitat types (such as rock, snow, vegetation, etc.) so that pixels are more clearly separated in the new subspace. This differentiation helps subsequent clustering or classification processes, enabling the model to classify each data point more accurately, and ultimately determine the habitat type to which the data point belongs by comparing the distance between pixel features and cluster centers.

[0107] Here, α is the control weight, used to balance the inter-class divergence objective and the reconstruction error. The larger the value, the more emphasis is placed on reconstruction consistency.

[0108] Specifically, the model calculation parameters of the preset animal density model mentioned in step S400 are confirmed in the following way:

[0109] S410: Obtain several habitat area image samples; perform adjustment processing on each habitat area image sample to obtain several computational image samples; the adjustment processing includes data cleaning, missing value handling and normalization processing.

[0110] The image samples are adjusted to reduce interference from factors such as noise and missing data, and the output image samples are used as the basis for subsequent calculations.

[0111] S420: Determine the population size and habitat type information corresponding to each computational image sample based on each computational image sample.

[0112] Based on the data information of the image samples, determine the population size information (population size distribution and quantity) and the distribution of habitat regional characteristic categories corresponding to the images.

[0113] S430: Perform pixel-by-pixel calculations based on each computed image sample and the population size information of each computed image sample to determine the population density information corresponding to each pixel unit of each computed image sample.

[0114] For each pixel, the population density information corresponding to the image sample is calculated, which serves as the data basis for subsequent regression calculations.

[0115] S440: Perform regression calculations based on the habitat type information corresponding to each calculated image sample and the population density information corresponding to each pixel unit of each calculated image sample, and determine the model calculation parameters of the preset animal density model based on the regression calculation results.

[0116] Regression calculations are performed based on population density data and habitat area feature category data from habitat images to construct a pre-defined animal density model (i.e., a population regression model) that correlates population density with habitat image feature categories. The model calculation parameters of the pre-defined animal density model are then determined based on the regression calculation results.

[0117] In one embodiment, penguins are selected as the target animal population; then, the animal density model relating penguin population density to habitat image feature categories is constructed as follows:

[0118] Based on the habitat classification results, regression models were constructed to connect the number of adult and juvenile penguins with the habitat image feature categories. Here, y represents penguin density, and x represents the habitat image feature category (X ranges from 1 to 4, corresponding to the outermost level of category 1 and the center of category 4). First, the habitats were divided into four categories, and then the penguin density corresponding to each category was determined. That is, x represents the habitat image feature category, and y represents the penguin density. An example of the specific calculation process for determining the animal density model is as follows:

[0119] We performed calculations on an island where the penguin population was previously determined. The processed image of this island has a unit pixel area of ​​2×2 m². We selected 10 units for each calculation, including the penguin density (adult and juvenile penguins were calculated separately). This calculation was repeated for each habitat to obtain the corresponding density. This establishes the relationship between X and Y, which is then used for regression analysis, resulting in the following regression model:

[0120] The density model for adult penguins is: y = –0.0018x + 0.471;

[0121] The density model for juvenile penguins is: y = –0.0013x + 0.399.

[0122] Furthermore, the animal population size prediction method provided in this embodiment of the invention also includes:

[0123] S600: Generates a trend map based on the predicted animal population size and spatial distribution images of the target animal population at different times.

[0124] The total population size, population size distribution, and spatial distribution images of the target animal population at different times are analyzed to generate trend maps, which serve as the data basis for subsequent predictions of population changes of the target animal.

[0125] S700: Based on the trend chart analysis, the predicted results of the population size changes of the target animal population at different times are obtained.

[0126] Analysis of trend charts determines the predicted population changes of target animal species at different times. Based on these results, analysis reveals population trends, growth rates or decline rates, population density, habitat suitability, and driving factors of population fluctuations. Furthermore, it includes extinction risk assessment, population recovery potential, and ecological impact assessment.

[0127] Implementing the embodiments of the present invention has the following beneficial effects:

[0128] (1) Feature extraction of images based on sparse feature selection method: Based on the preset sparse feature selection matrix, feature selection is performed on the feature matrix of each pixel in the image to retain the features that are more helpful for classification in complex images and improve the accuracy of subsequent habitat type prediction.

[0129] (2) Determine the similarity between data points and cluster centers based on loss function: In the method of this invention, several initial cluster centers are randomly selected, and the similarity between data points and cluster centers is calculated based on the loss function and the distance between data points and cluster centers to reduce the impact of some abnormal interference in the image on the splitting results.

[0130] (3) Feature category clustering based on alternating optimization method: In each step, the classification level weights and cluster center positions of the data points are readjusted according to the current classification results, and then the classification labels of each data point are updated. This process will continue to cycle until the feature category clustering results of each data point are relatively stable and the overall classification effect is optimal; thus optimizing the feature category clustering effect can improve the accuracy of predicting habitat species using complex images.

[0131] like Figure 3 As shown, this embodiment of the invention also provides an animal population size prediction system, including:

[0132] The feature extraction module is used to acquire remote sensing images of the animal habitats of the target animal population in the target area; and to extract features from the feature information of each pixel in the animal habitat remote sensing image to obtain the feature matrix corresponding to each pixel.

[0133] The feature selection module is used to select features from the feature matrix corresponding to each pixel based on the sparse feature selection method, and construct a simplified matrix containing several data points based on the feature selection results; the feature information of several data points is determined by the feature selection results of the feature matrix corresponding to several pixels.

[0134] The classification prediction module is used to select several cluster centers from the simplified matrix; based on the selection of several cluster centers, the feature information on several data points and the preset classification model, the feature categories are clustered using an alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image;

[0135] Density prediction module: Calculates the population density corresponding to each habitat area image based on several habitat area images and a preset animal density model;

[0136] The analysis module is used to analyze the habitat species and population density corresponding to several habitat area images to determine the predicted animal population size of the target animal species.

[0137] It is evident that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0138] like Figure 4 As shown, this embodiment of the invention also provides an animal population size prediction device, comprising:

[0139] At least one processor;

[0140] At least one memory for storing at least one program;

[0141] When the at least one program is executed by the at least one processor, the at least one processor implements the steps of the animal population size prediction method described in the above method embodiments.

[0142] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. The memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include remote memory located remotely relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0143] It is evident that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented in the present device embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0144] Furthermore, embodiments of this application also disclose a computer program product or computer program stored in a computer-readable storage medium. A processor of a computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, causing the computer device to perform the methods described above.

[0145] This invention also provides a computer-readable storage medium storing a processor-executable program that, when executed by a processor, implements the above-described method. Similarly, the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0146] It is understood that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0147] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for predicting animal population size, characterized in that, include: Acquire remote sensing images of the animal habitat of the target animal population in the target area; extract features from the feature information of each pixel in the animal habitat remote sensing image to obtain a feature matrix corresponding to each pixel; Based on the sparse feature selection method, feature selection is performed on the feature matrix corresponding to each pixel, and a simplified matrix containing several data points is constructed according to the feature selection result; the feature information on the several data points is determined by the feature selection result of the feature matrix corresponding to the several pixels. Several cluster centers are selected from the simplified matrix; based on the selection of several cluster centers, the feature information on several data points, and the preset classification model, feature category clustering is performed based on an alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image; The population density corresponding to each habitat area image is calculated based on each habitat area image, its corresponding habitat species, and a preset animal density model. The population size prediction results of the target animal species are determined by analyzing the habitat species and population density corresponding to several habitat area images.

2. The method according to claim 1, characterized in that, The feature information of the pixel includes its spectral reflectance values ​​in several spectral bands; the simplified matrix is ​​constructed in the following way: Based on the spectral reflectance values ​​of each pixel in several bands, the band signal-to-noise ratio of each pixel in each band is calculated, and the auxiliary weight vector corresponding to each pixel is determined according to the calculation results. Based on a preset sparse feature matrix and an auxiliary weight vector corresponding to each pixel, feature selection is performed on the feature matrices of several pixels to obtain a simplified feature matrix corresponding to each pixel. Construct a simplified matrix containing several data points; the feature information on the several data points corresponds one-to-one with the simplified feature matrix corresponding to the several pixels.

3. The method according to claim 2, characterized in that, The expression for the preset sparse feature matrix is ​​as follows: Among them, s T (diag(M)+αw) is the evaluation function of the feature matrix corresponding to the pixel; diag(M) is the importance measure of each feature in the feature matrix corresponding to the pixel; α is the weight coefficient, w is the auxiliary weight vector; d represents the number of features selected from the feature matrix corresponding to the pixel.

4. The method according to claim 1, characterized in that, The feature information on several data points includes the feature information on several clustering center points; the step of clustering feature categories based on the selection of several clustering center points, the feature information on several data points, and a preset classification model, using an alternating optimization method, to obtain several habitat area images and the habitat type information corresponding to each habitat area image, includes: Based on a preset loss function, the selection of several cluster centers, the feature information of several data points, the feature information of several cluster centers, and a preset classification model, feature category clustering is performed based on an alternating optimization method to determine the habitat type corresponding to each data point. The pixels corresponding to data points of the same habitat species are merged according to their spatial location to obtain several habitat area images and their corresponding habitat species.

5. The method according to claim 4, characterized in that, The habitat type corresponding to each pixel is determined in the following way: Based on the feature information of the data points and the feature information of each cluster center point, the distance between the data points and each cluster center point is calculated; the distance calculation method includes Euclidean distance algorithm and cosine similarity algorithm. Set the initial classification level weight of the data points; calculate the similarity between the data points and each cluster center point based on the preset loss function, the initial classification level weight of the data points, and the distance between the data points and each cluster center point; Based on the similarity between the data points and each cluster center, the data points are clustered by feature category to determine the feature category clustering result of the data points; Adjust the selection of several cluster center points; Based on a preset formula, each data point is re-clustered according to its feature category, and the classification level weight of each data point is adjusted according to the feature category clustering results of each data point until the feature category clustering results of each data point tend to be stable. The habitat type corresponding to the data point is determined according to the adjusted feature category clustering results of the data points.

6. The method according to claim 5, characterized in that, The expression for the preset loss function is: Where tensor A represents the reconstruction error matrix, which is determined by the distance between the i-th data point and each cluster center point, ||A|| σ γL represents the weighted norm result of tensor A under the robust adjustment factor σ; i L represents the classification rank weight of the i-th data point, γ is the global scaling factor, and L i It is a local ranking factor; α i The eigenvector of the i-th data point is determined by the feature information of the data point; ||α i ||2 is the vector magnitude of the feature vector of the i-th data point; Let L2 be the squared L2 norm of the eigenvector of the i-th data point.

7. The method according to claim 5, characterized in that, The expression of the preset formula is: max W,F,G Tr(W T S t W)-α||Γ 1 / 2 (X T W-FG T )|| σ ; in, Let Tr(W) be the optimization function for vectors W, F, and G, where W represents the feature matrix of the data points, F represents the clustering result of the feature categories of the data points, and G represents the selection of several cluster centers; T S t W) is the inter-class divergence target, determined by the degree of distribution difference in the clustering results of the corresponding feature categories for each data point; S t The covariance matrix measures the degree of difference in the distribution of clustering results for the corresponding feature categories of each data point; (X T W-FG T ) represents the reconstruction error, determined by the distance between data points and each cluster center; α is the control weight, used to balance the inter-class divergence objective and the reconstruction error; Γ 1 / 2 This is a weight matrix used to represent the importance of the data points in the loss function.

8. The method according to claim 1, characterized in that, The model calculation parameters of the preset animal density model are confirmed in the following way: Acquire several habitat area image samples; perform adjustment processing on each of the habitat area image samples to obtain several computational image samples; the adjustment processing includes data cleaning, missing value handling, and normalization processing; Based on each computational image sample, determine the population size information and habitat type information corresponding to each computational image sample; Based on each computational image sample and the population size information of each computational image sample, pixel-by-pixel calculations are performed to determine the population density information corresponding to each pixel unit of each computational image sample. Regression calculations are performed based on the habitat type information corresponding to each calculated image sample and the population density information corresponding to each pixel unit of each calculated image sample. The model calculation parameters of the preset animal density model are determined based on the regression calculation results.

9. An animal population size prediction system, characterized in that, include: The feature extraction module is used to acquire remote sensing images of the animal habitat of the target animal population in the target area; and to extract features from the feature information of each pixel in the remote sensing image of the animal habitat to obtain a feature matrix corresponding to each pixel. The feature selection module is used to perform feature selection on the feature matrix corresponding to each pixel based on the sparse feature selection method, and construct a simplified matrix containing several data points based on the feature selection result; the feature information on the several data points is determined by the feature selection result of the feature matrix corresponding to the several pixels; The classification prediction module is used to select several cluster centers from the simplified matrix; based on the selection of several cluster centers, the feature information on several data points and the preset classification model, feature category clustering is performed based on an alternating optimization method to obtain several habitat area images and the habitat type information corresponding to each habitat area image; Density prediction module: used to calculate the population density corresponding to each habitat area image based on each habitat area image, its corresponding habitat species, and a preset animal density model; The analysis module is used to analyze the habitat species and population density corresponding to several habitat area images to determine the predicted animal population size of the target animal population.

10. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the method as described in any one of claims 1 to 8.

Images