Data processing system and method for biodiversity monitoring

Abstract

The invention relates to the field of biodiversity monitoring, and particularly discloses a data processing system and method for biodiversity monitoring, and the method comprises the steps: firstly obtaining an earth surface coverage image collected by a high-resolution remote sensing image, environment sound information collected by a sound sensor, and biological survey data collected by a database; and then feature extraction and correlation analysis are performed on the three by using a deep learning technology, and finally a biodiversity heat map is generated through a generator, so that the biodiversity monitoring efficiency and accuracy are improved, and the understanding of the health degree of an ecological system is facilitated.

Description

Technical Field

[0001] The present application relates to the field of biodiversity monitoring, and more specifically, to a data processing system and method for biodiversity monitoring. Background Art

[0002] Population ecology originated from fields such as demography and focuses on studying the changes in population size and its interaction with the environment. This discipline has been widely used in environmental assessment, resource development and management, and disaster prevention and control. Competition between populations and the relationship between predators and prey are basic forms of their interaction and are also important research topics in the fields of ecology and biomathematics. The phenomenon of predation and prey is ubiquitous in nature, and this relationship forms the basis of biological communities. The food chain is a specific manifestation of this relationship, describing the process of chemical energy being transferred layer by layer between organic matter in an ecosystem.

[0003] In existing biodiversity monitoring methods, analysis is often conducted only from a single data source, which can easily lead to information fragmentation and misunderstanding.

[0004] Therefore, a data processing system and method for biodiversity monitoring is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a data processing system and method for biodiversity monitoring, which first obtains a surface cover image collected by a high-resolution remote sensing image, environmental sound information collected by a sound sensor, and biological survey data collected by a database, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally generates a biodiversity heat map through a generator, thereby improving the efficiency and accuracy of biodiversity monitoring and helping to understand the health of the ecosystem.

[0006] According to one aspect of the present application, there is provided a data processing system for biodiversity monitoring, comprising: Biodiversity monitoring data acquisition module, used to obtain land cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases; A biodiversity monitoring data processing module, used to extract biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database; The biodiversity heat map generation module is used to generate a biodiversity heat map based on the biological multimodal information association feature vector and the biological survey data global feature vector.

[0007] According to another aspect of the present application, a data processing method for biodiversity monitoring is provided, comprising: Obtain land cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases; Extracting biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database; A biodiversity heat map is generated based on the biological multimodal information association feature vector and the biological survey data global feature vector.

[0008] Compared with the prior art, the present application provides a data processing system and method for biodiversity monitoring, which first obtains surface cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally generates a biodiversity heat map through a generator, thereby improving the efficiency and accuracy of biodiversity monitoring and helping to understand the health of the ecosystem. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 4 is a block diagram of a data processing system for biodiversity monitoring according to an embodiment of the present application.

[0011] Figure 2 4 is a block diagram of a biodiversity monitoring data processing module in a data processing system for biodiversity monitoring according to an embodiment of the present application.

[0012] Figure 3 It is a block diagram of a ground cover image feature extraction unit in a data processing system for biodiversity monitoring according to an embodiment of the present application.

[0013] Figure 44 is a block diagram of a biodiversity heat map generation module in a data processing system for biodiversity monitoring according to an embodiment of the present application.

[0014] Figure 5 Flow chart of a data processing method for biodiversity monitoring according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0016] Figure 1 FIG. 1 is a block diagram of a data processing system for biodiversity monitoring according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, the data processing system 100 for biodiversity monitoring includes: a biodiversity monitoring data acquisition module 110, which is used to acquire a surface cover image collected by a high-resolution remote sensing image, environmental sound information collected by a sound sensor, and biological survey data collected by a database; a biodiversity monitoring data processing module 120, which is used to extract biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database; a biodiversity heat map generation module 130, which is used to generate a biodiversity heat map based on the biological multimodal information association feature vectors and the biological survey data global feature vectors.

[0017] In the above-mentioned data processing system 100 for biodiversity monitoring, the biodiversity monitoring data acquisition module 110 is used to obtain surface cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases. It should be understood that population ecology originally originated from related fields such as demography, and mainly focuses on changes in population size and its relationship with the environment. This discipline has been widely used in many fields, including environmental assessment, resource development and management, and disaster prevention and control. Among them, the competitive relationship between populations and the interaction between predators and prey constitute important research topics in ecology and biomathematics. In nature, the phenomenon of predation and prey is ubiquitous, forming the basic structure of the ecosystem. Among them, the food chain is the embodiment of this relationship, depicting how chemical energy is transmitted step by step between organisms in the ecosystem. However, many current biodiversity monitoring methods often rely on a single data source for research, which is prone to one-sided information and misunderstanding. Therefore, in the technical solution of the present application, surface cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases are obtained, and combined with deep learning technology, by generating biodiversity heat maps, the efficiency and accuracy of biodiversity monitoring are significantly improved, so as to deeply understand the health status of the ecosystem.

[0018] Specifically, high-resolution remote sensing images provide detailed surface cover information, which can effectively identify different ecological regions and their changes, such as forests, wetlands and the impact of urbanization. At the same time, the environmental sound information collected by sound sensors adds an important dimension to ecological monitoring. Sound is a key indicator of biological interactions in ecosystems, which can reflect animal activities, population density and biodiversity. The biological survey data obtained from the database provides important information on species distribution, population size and ecological habits. Combining these data sources, multimodal information for biodiversity monitoring can be obtained, thereby enhancing the accuracy and depth of the analysis. This comprehensive approach can not only reduce the fragmentation of information, but also reveal the relationship between different data and provide a more comprehensive ecological assessment.

[0019] In the above-mentioned data processing system 100 for biodiversity monitoring, the biodiversity monitoring data processing module 120 is used to extract biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database. It should be understood that the extraction and association of such multimodal information provides a powerful tool for the overall understanding and management of the ecosystem, helps to meet the challenges of biodiversity decline, and promotes the goal of sustainable development. Specifically, combining the advantages of different data sources will bring deeper insights to scientific research and promote the effective implementation of ecological protection.

[0020] Figure 2 FIG. 4 is a block diagram of a biodiversity monitoring data processing module in a data processing system for biodiversity monitoring according to an embodiment of the present application. Figure 2 As shown, in a specific embodiment of the present application, the biodiversity monitoring data processing module 120 includes: a surface cover image feature extraction unit 121, used to perform feature extraction on the surface cover image collected by the high-resolution remote sensing image to obtain a surface cover feature vector of interest; an environmental sound information feature extraction unit 122, used to perform feature extraction on the environmental sound information collected by the sound sensor to obtain an environmental sound audio global feature vector; a biological multimodal information feature association unit 123, used to associate the surface cover feature vector of interest with the environmental sound audio global feature vector to obtain the biological multimodal information association feature vector; a biological survey data feature extraction unit 124, used to perform feature extraction on the biological survey data collected by the database to obtain the biological survey data global feature vector.

[0021] It should be understood that high-resolution remote sensing images can capture subtle surface features, such as vegetation type, land use, urban expansion, etc. This information is crucial for assessing the health of ecosystems. However, in high-resolution images, the amount of data is usually very large, and direct analysis may lead to excessive computational burden. Through feature extraction, these complex image data can be converted into concise and meaningful feature vectors, which can help researchers identify specific ecological regions and their characteristics, thereby analyzing the impact of these regions on biodiversity. In addition, the data dimension is effectively reduced while retaining the most important information. This method not only improves computational efficiency, but also reduces noise interference, making the analysis results more reliable.

[0022] Furthermore, environmental sound information contains rich biological and environmental data. These sounds include not only the calls of animals such as birds and insects, but may also include the sounds of wind, water flow and human activities. Feature extraction can convert raw audio signals into numerical feature vectors, which can more effectively describe the characteristics of sound, such as the frequency, loudness, duration and sound quality of the audio. This data preprocessing is the basis for analyzing and understanding audio data, and can significantly improve the accuracy and efficiency of subsequent analysis. By extracting global audio features, the state of biodiversity and its changes in the ecosystem can be revealed. For example, the sound characteristics of different species are unique. By analyzing features such as frequency and loudness, the presence and reproduction of specific species can be monitored, and the health of the biological community can be assessed.

[0023] Furthermore, the surface cover feature vector provides static information about the ecological environment, including the distribution and changes of different surface types (such as forests, grasslands, water bodies, etc.). This information helps to understand the habitat preferences and survival conditions of various organisms under specific environmental conditions. The global feature vector of environmental sound audio reflects dynamic information related to biological activities, including the calls, communication and activity of organisms. Correlating these two types of information can reveal how organisms move and adapt under different environmental conditions. Specifically, by integrating information from different data sources, researchers can establish more comprehensive and accurate ecological models to explain and predict biodiversity changes and ecological risks. This model not only improves the reliability of the analysis, but also provides a scientific basis for environmental management.

[0024] In particular, biological survey data usually contain a lot of information about species distribution, population size, ecological characteristics and habitat requirements. The global feature vector of biological survey data can reflect the overall characteristics of the ecosystem, such as species diversity, community structure and ecological function. By statistically analyzing the richness and relative abundance of different species and their association with environmental factors, researchers can identify the health status of the ecosystem and changes in biodiversity. These global features are of great reference value in assessing ecosystem service functions, the impact of environmental changes and formulating conservation measures.

[0025] Figure 3 FIG. 1 is a block diagram of a land cover image feature extraction unit in a data processing system for biodiversity monitoring according to an embodiment of the present application. Figure 3As shown, in a specific embodiment of the present application, the surface cover image feature extraction unit 121 includes: a surface cover image target detection subunit 1211, which is used to perform target detection on the surface cover image collected by the high-resolution remote sensing image to obtain a surface cover area of ​​interest; a surface cover significant convolution coding subunit 1212, which is used to pass the surface cover area of ​​interest through a surface cover significant target detector of interest to obtain a surface cover feature map of interest; a surface cover feature of interest pooling subunit 1213, which is used to perform average value pooling on the surface cover feature map of interest based on a feature matrix with a preset size to obtain a surface cover feature vector of interest.

[0026] It should be understood that high-resolution remote sensing images have strong spatial resolution and can accurately display surface features, such as different types of surface cover such as urban buildings, farmland, water bodies, forests and grasslands. In complex natural environments, surface features are often confused. For example, an urban area and its adjacent green space may have similar color patterns in the image, which may lead to classification errors. By introducing a deep learning algorithm for target detection and combining the contextual information in high-resolution images, different types of surface cover can be more accurately distinguished, thereby improving the credibility of the classification results. Specifically, the surface cover image collected by the high-resolution remote sensing image is processed using the following target detection formula to obtain the surface cover area of ​​interest; Among them, the target detection formula is: ;in, It is a land cover image collected by high-resolution remote sensing images. is the anchor frame, For the surface coverage area of ​​interest, Indicates classification, Indicates regression.

[0027] Furthermore, salient object detectors are specifically designed to identify and extract salient features and objects in images. In the context of land cover, these objects may include different types of land cover, such as urban buildings, farmland, forests, wetlands, etc. By applying these detectors, representative feature areas can be efficiently extracted from high-resolution remote sensing images to generate land cover feature maps of interest. This automated process greatly reduces the time and effort of manual analysis and improves the overall processing efficiency. Moreover, considering the background of rapid urbanization and climate change, changes in land cover types directly affect ecological balance and biodiversity. By using the feature maps extracted by salient object detectors, researchers can accurately track and analyze land surface change trends and identify potential ecological risks and problems in a timely manner. This is of great significance for the formulation of environmental protection policies and the implementation of sustainable management. Specifically, each layer of the ground cover salient object of interest detector performs the following on the input data in the forward pass of the layer: uses a first convolution kernel to convolve the input data to obtain a first convolution feature map; uses a second convolution kernel to convolve the first convolution feature map to obtain a second convolution feature map, wherein the size of the first convolution kernel is larger than the size of the second convolution kernel; and, performs pooling on the second convolution feature map to obtain a pooled feature map; and, performs activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the ground cover salient object of interest detector is the ground cover feature map of interest, and the input of the first layer of the ground cover salient object of interest detector is the ground cover region of interest.

[0028] Furthermore, considering that the land cover feature map of interest usually contains high-dimensional information, including pixel values ​​and corresponding features at multiple spatial locations. Although this information is rich, if it is not processed, it may lead to increased computational complexity and storage requirements. By performing mean pooling on the land cover feature map of interest based on a feature matrix with a preset size, the local area information in the feature map can be effectively integrated into a representative mean, thereby reducing the dimension of the data. This dimensionality reduction process not only reduces the computational burden, but also speeds up subsequent analysis, especially when processing large-scale remote sensing data. Specifically, in the pooling operation, each area of ​​a preset size in the feature map is converted into a mean, which means that there is a certain inhibitory effect on noise and changes in a small range. For example, some minor changes or local interference in the image will not have a significant impact on the final feature vector, making the extracted feature vector more representative and stable. The feature vector generated after mean pooling can be combined with other data sources (such as environmental factors, climate data, etc.) to support more comprehensive ecological research and decision support. Through this multi-dimensional data fusion, researchers can obtain deeper insights, including identifying key environmental changes and evaluating ecosystem services.

[0029] In a specific embodiment of the present application, the ambient sound information feature extraction unit 122 includes: extracting Mel-frequency cepstral coefficients from the ambient sound information collected by the sound sensor to obtain multiple ambient sound information Mel-frequency cepstral coefficient short-time eigenvalues; passing the multiple ambient sound information Mel-frequency cepstral coefficient short-time eigenvalues ​​through an ambient sound information Mel-frequency cepstral feature encoder based on a convolutional neural network model to obtain the ambient sound audio global feature vector.

[0030] It should be understood that Mel-frequency cepstral coefficients (MFCC) are a feature extraction method that simulates the way the human ear perceives sound signals. The human ear's perception of sounds in different frequency bands is not linear, and low-frequency sounds are more sensitive than high-frequency sounds. The Mel frequency scale simulates this perceptual characteristic. By converting the frequency domain signal to the Mel scale, the characteristics of the ambient sound can be more naturally reflected. This conversion enables MFCC to capture important acoustic features in the audio signal, which is helpful for subsequent analysis and recognition. Considering that the ambient sound signal usually contains a large amount of raw data, directly processing these high-dimensional data is not only computationally complex, but may also lead to problems such as overfitting. By extracting MFCC, the high-dimensional sound signal can be converted into a low-dimensional feature representation, thereby reducing the amount of data and reducing the computational burden. This feature dimensionality reduction process can improve the efficiency of the algorithm, especially in real-time processing and large-scale data analysis. Among them, by processing the signal in frames and extracting MFCC for each frame, the dynamic changes in the spectrum of the sound signal in a short period of time can be obtained, so as to better understand the law of sound changes. This short-term analysis capability is particularly important in applications such as speech recognition, acoustic event monitoring, and ambient sound classification.

[0031] Furthermore, MFCC short-time feature values ​​capture the time-varying characteristics of sound signals through frame processing, but these short-time features often only describe local information of the sound and lack understanding of global features. Convolutional neural network (CNN) is a deep learning model that is good at processing local features and can extract spatial features from data through convolution layers. When the MFCC short-time feature values ​​are input into the CNN model, the model can automatically extract and learn local features through a combination of multiple convolutional layers and pooling layers, and finally obtain a global feature vector. This method enables the model to extract deeper features from complex sound signals, thereby improving the effect of acoustic analysis. Specifically, the short-time eigenvalues ​​of the Mel-frequency cepstral coefficients of the multiple ambient sound information are converted into a sound input vector; each layer of the ambient sound information Mel-frequency cepstral feature encoder based on the convolutional neural network model is used to perform the following operations on the input data in the forward transfer of the layer: convolution processing based on a one-dimensional convolution kernel is performed on the input data to obtain a convolution feature map; pooling processing based on a feature matrix is ​​performed on the convolution feature map to obtain a pooled feature map; and nonlinear activation is performed on the pooled feature map to obtain an activation feature map; wherein the input of the first layer of the ambient sound information Mel-frequency cepstral feature encoder based on the convolutional neural network model is the sound input vector, and the output of the last layer of the ambient sound information Mel-frequency cepstral feature encoder based on the convolutional neural network model is the ambient sound audio global feature vector.

[0032] In a specific embodiment of the present application, the biological multimodal information feature association unit 123 includes: adding a dependency library through Maven; using DruidDataSource to configure the data source, including database connection information, connection pool size, and connection timeout; when the application starts, initializing DruidDataSource and setting it as a globally accessible data source; obtaining a surface coverage feature vector of interest and a global feature vector of ambient sound audio; receiving two feature vectors as input in calculateMultiModalFeatureVector, calculating the product of the two feature vectors, obtaining a biological multimodal information association feature vector, and returning the calculation result; and storing the calculated biological multimodal information association feature vector back in the database.

[0033] Specifically, some deployment codes are as follows: import DruidDataSource; import java.sql.Connection; import java.sql.PreparedStatement; import java.sql.ResultSet; import java.sql.SQLException; public class MultiModalFeatureCalculator { private static DruidDataSource dataSource; public static void main(String[] args) { / / Initialize the data source initDataSource(); / / Get the eigenvector and calculate the product try { double[] surfaceFeatureVector = getSurfaceFeatureVector(); double[] audioFeatureVector = getAudioFeatureVector(); double[] multiModalFeatureVector = calculateMultiModalFeatureVector(surfaceFeatureVector, audioFeatureVector); / / Store or output the result storeResult(multiModalFeatureVector); } catch (SQLException e) { e.printStackTrace(); } finally { / / Close the data source closeDataSource(); } } private static void initDataSource() { dataSource = new DruidDataSource(); dataSource.setUrl("jdbc:mysql: / / localhost:3306 / your_database"); dataSource.setUsername("your_username"); dataSource.setPassword("your_password"); dataSource.setInitialSize(5); dataSource.setMaxActive(10); } private static double[] getSurfaceFeatureVector() throws SQLException { String sql = "SELECT feature_vector FROM surface_features WHERE id =?"; try (Connection conn = dataSource.getConnection(); PreparedStatement pstmt = conn.prepareStatement(sql)) { pstmt.setInt(1, 1); / / id is 1 ResultSet rs = pstmt.executeQuery(); if (rs.next()) { String featureVectorStr = rs.getString("feature_vector"); return parseFeatureVector(featureVectorStr); } } return null; } private static double[] getAudioFeatureVector() throws SQLException{ String sql = "SELECT feature_vector FROM audio_features WHEREid = ?"; try (Connection conn = dataSource.getConnection(); PreparedStatement pstmt = conn.prepareStatement(sql)) { pstmt.setInt(1, 1); / / id is 1 ResultSet rs = pstmt.executeQuery(); if (rs.next()) { String featureVectorStr = rs.getString("feature_vector"); return parseFeatureVector(featureVectorStr); } } return null; } private static double[] parseFeatureVector(String featureVectorStr) { String[] parts = featureVectorStr.split(","); double[] featureVector = new double[parts.length]; for (int i = 0; i<parts.length; i++) { featureVector[i] = Double.parseDouble(parts[i]); } return featureVector; } private static double[] calculateMultiModalFeatureVector(double[]surfaceFeatureVector, double[] audioFeatureVector) { if (surfaceFeatureVector.length != audioFeatureVector.length) { throw new IllegalArgumentException("Feature vectors must have thesame length"); } double[] result = new double[surfaceFeatureVector.length]; for (int i = 0; i<surfaceFeatureVector.length; i++) { result[i] = surfaceFeatureVector[i]* audioFeatureVector[i]; } return result; } private static void storeResult(double[] multiModalFeatureVector) throws SQLException { String sql = "INSERT INTO multi_modal_features (feature_vector) VALUES (?)"; try (Connection conn = dataSource.getConnection(); PreparedStatement pstmt = conn.prepareStatement(sql)) { pstmt.setString(1, arrayToString(multiModalFeatureVector)); pstmt.executeUpdate(); } } private static String arrayToString(double[] array) { StringBuilder sb = new StringBuilder(); for (double value : array) { sb.append(value).append(","); } sb.deleteCharAt(sb.length() - 1); / / Delete the last comma return sb.toString(); } private static void closeDataSource() { if (dataSource != null) { dataSource.close(); } }

[0034] Among them, in the data source initialization, the DruidDataSource is configured in the initDataSource() method, including the database connection URL, username, password, and the initial and maximum sizes of the connection pool. In feature vector acquisition, the getSurfaceFeatureVector() and getAudioFeatureVector() methods obtain the surface cover feature vector of interest and the global feature vector of ambient sound audio from the database respectively. In feature vector calculation, the calculateMultiModalFeatureVector() method calculates the product of two feature vectors to obtain the biological multimodal information association feature vector. In result storage, the storeResult() method stores the calculation results back to the database. In resource closing, the closeDataSource() method closes the data source and releases resources at the end of the program.

[0035] Specifically, use the DruidDataSource class to initialize the data source, configure the database connection URL, user name, password, and the initial and maximum size of the connection pool. The specific implementation method is named initDataSource().

[0036] Get the surface coverage feature vector of interest. The specific implementation method is named getSurfaceFeatureVector(). This method obtains the feature vector from the database by executing a SQL query statement and parses it into a double-precision floating-point array.

[0037] Get the global feature vector of ambient sound audio. The specific implementation method is named getAudioFeatureVector(). This method obtains the feature vector from the database by executing SQL query statements and parses it into a double-precision floating-point array. Calculate the product of the feature vector of interest of surface cover and the global feature vector of ambient sound audio to obtain the feature vector associated with biological multimodal information. The specific implementation method is named calculateMultiModalFeatureVector(). This method receives two feature vectors as input and returns the calculation result.

[0038] The calculated biological multimodal information associated feature vector is stored back in the database. The specific implementation method is named storeResult(). This method stores the feature vector in the database by executing an SQL insert statement. At the end of the program, the data source is closed to release resources. The specific implementation method is named closeDataSource().

[0039] It should be understood that by dynamically managing database connections, Druid can significantly reduce the overhead of connection creation and destruction, and improve the performance and stability of database operations. Among them, Druid is an open source distributed data storage system designed for real-time analysis and big data processing. It combines the characteristics of data warehouses, time series databases, and search systems, and can quickly load and query large-scale data sets. Druid is particularly suitable for scenarios that require low latency and high concurrent queries, such as real-time monitoring, advertising analysis, and user behavior analysis.

[0040] In this way, the calculation of biological multimodal information association feature vectors is realized, and the overall performance and reliability of the system are improved by optimizing database operations, ensuring data consistency, and facilitating maintenance and resource management.

[0041] In a specific embodiment of the present application, the biological survey data feature extraction unit 124 includes: passing the biological survey data collected from the database through a biological survey data text semantic encoding model to obtain multiple biological survey data feature vectors; and splicing the multiple biological survey data feature vectors into the biological survey data global feature vector.

[0042] It should be understood that biological survey data usually exist in the form of text, containing rich descriptive information, such as species characteristics, ecological environment, distribution, etc. The complexity of these text data makes direct analysis and processing difficult. Text semantic encoding models, especially methods based on deep learning (such as BERT, Word2Vec, etc.), can convert these text data into numerical feature vectors. These vectors can not only capture the semantic information of the text, but also reflect the contextual relationship between words, making the processing results more informative and representative. Specifically, the biological survey data collected from the database are segmented to obtain a biological survey word sequence; the embedding layer of the biological survey data text semantic encoding model is used to map each biological survey word in the biological survey word sequence to a word embedding vector to obtain a sequence of biological survey word embedding vectors; the converter-based Bert model of the biological survey data text semantic encoding model is used to perform global context semantic encoding on the sequence of biological survey word embedding vectors to obtain multiple biological survey data feature vectors.

[0043] Furthermore, biological survey data usually come from multiple dimensions and features, including ecological characteristics, environmental impacts, learning samples, etc. These features are often represented in vector form, and each feature vector captures a specific information dimension. If only a certain feature vector is used alone, other important information may be ignored, resulting in one-sided analysis. By splicing multiple feature vectors, a more complete global feature vector can be obtained, which integrates various important information and makes subsequent analysis more comprehensive.

[0044] In the above-mentioned data processing system 100 for biodiversity monitoring, the biodiversity heat map generation module 130 is used to generate a biodiversity heat map based on the biological multimodal information association feature vector and the biological survey data global feature vector. It should be understood that biological multimodal information includes data from different sources and types, such as images, texts, environmental data, etc., while biological survey data provides specific ecological characteristics. The combination of these feature vectors enables researchers to analyze biodiversity from multiple dimensions to fully capture the relationships between species and the response to environmental changes, thereby forming a deeper ecological understanding. The process of generating a heat map can effectively display the spatial distribution of biodiversity. Specifically, the heat map represents the density or distribution of biodiversity through color gradients, allowing researchers to quickly identify areas with rich or poor diversity. This intuitive visualization form can help ecologists, decision makers and the public quickly obtain important information, thereby providing a basis for ecological protection and biodiversity management.

[0045] Figure 4 FIG. 4 is a block diagram of a biodiversity heat map generation module in a data processing system for biodiversity monitoring according to an embodiment of the present application. Figure 4 As shown, in a specific embodiment of the present application, the biodiversity heat map generation module 130 includes: a biodiversity data feature fusion unit 131, which is used to fuse the biological multimodal information association feature vector and the biological survey data global feature vector to obtain a biological heat map generation feature vector; a biodiversity data feature optimization unit 132, which is used to maximize the entropy degradation of the biological heat map generation feature vector based on the confusion of the learning parameter's own structural information to obtain a sparse optimized biological heat map generation feature vector; a biodiversity data feature generation unit 133, which is used to pass the sparse optimized biological heat map generation feature vector through a generator to generate a biodiversity heat map.

[0046] It should be understood that by integrating data from different sources, the model can learn more complex patterns, thereby improving the prediction ability of biodiversity. This approach not only helps to improve the quality of heat map generation, but also helps to identify key ecological drivers and promote the understanding of ecosystem changes.

[0047] In particular, in the technical solution of the present application, the feature vector generated by the bioheat map usually contains multiple features, reflecting the multimodal information extracted from high-resolution remote sensing images, environmental sound information, and biological survey data. The dimensions and distribution of these features may be very complex and diverse. If the structure of the learning parameter matrix fails to effectively capture the association between these features, or the diversity and hierarchy of the features are not taken into account during the design, feature redundancy will result. For example, if certain nodes or channels in the learning parameter matrix correspond to the highly redundant parts of the features, then when performing feature fusion, these redundant information may be unnecessarily amplified, resulting in repeated or similar feature representations in the generated biodiversity heat map. Further, when the feature vector generated by the bioheat map passes through the learning parameter matrix, if the design of the matrix fails to accurately reflect the semantic relationship or spatial relationship between the features, it may cause information interference between the features. For example, the specific representation of some bioheat map generation feature vectors may overlap with the environmental sound features, but the learning parameter matrix fails to effectively distinguish these features, resulting in similar feature information being mixed in the final output. This confusion will not only affect the quality of the generated heat map, but may also cause the overfitting phenomenon of the model, that is, the model performs well on the training set, but cannot generalize on new data and cannot accurately reflect the real biodiversity situation. Therefore, in the technical solution of the present application, the entropy degradation maximization adjustment based on the confusion of the learning parameter's own structural information is performed on the biological heat map generation feature vector to obtain a sparse optimized biological heat map generation feature vector.

[0048] Among them, the entropy degradation maximization adjustment based on the confusion of the learning parameter's own structural information is performed on the biological heat map generation feature vector to obtain a sparse optimized biological heat map generation feature vector, including: extracting each row vector of the learning parameter matrix in the generator as a set of biological heat map learning parameter row vectors; calculating the biological heat map entropy degradation maximization simulation value between the biological heat map generation feature vector and each biological heat map learning parameter row vector in the set of biological heat map learning parameter row vectors to obtain a set of biological heat map entropy degradation maximization simulation values; based on the comparison between each biological heat map entropy degradation maximization simulation value in the set of biological heat map entropy degradation maximization simulation values ​​and a preset threshold, the learning parameter matrix is ​​gated and pruned to obtain a biological heat map learning parameter entropy degradation optimization matrix; the biological heat map generation feature vector is matrix-multiplied with the biological heat map learning parameter entropy degradation optimization matrix to obtain the sparse optimized biological heat map generation feature vector.

[0049] The optimization steps are specifically expressed as follows: ; ; ; ; ; ;in, is the set of row vectors of parameters for biological heatmap learning, is a set operation, , , , are the first, second, and third row vectors of the biological heat map learning parameters. , The eigenvalues ​​of the row vector of biological heatmap learning parameters, The first value in the set of simulation values ​​that maximizes the entropy degradation of the biological heat map The biological heat map entropy degradation maximizes the simulation value, is the maximum value function, Represents the calculation vector and The Euclidean distance between Generate feature vectors for biological heatmaps, For the The transposed vector of the row vector of biological heatmap learning parameters, is the length of the feature vector, is matrix multiplication, , , , The first, second, and , learning parameter mask values, For mask operation, Indicates the preset threshold value, Learning parameter entropy degradation optimization matrix for biological heatmaps, Generate feature vectors for sparse optimized biological heatmaps.

[0050] In the technical solution of the present application, it is taken into consideration that in the process of inputting the biological heat map generating feature vector into the learning parameter matrix of the generator to obtain the biodiversity heat map, there is a mismatch between the structure of the learning parameter matrix itself and the biological heat map generating feature vector, that is, the learning parameter matrix has structural redundancy and information confusion relative to the biological heat map generating feature vector, which may cause feature redundancy or noise in the biodiversity heat map, resulting in model overfitting.

[0051] Based on this, in the technical solution of the present application, the entropy degradation maximization adjustment based on the confusion of the learning parameter's own structural information is performed on the biological heat map generating feature vector, which uses the biological heat map generating feature vector as a wandering node, and uses the feature distribution between the biological heat map generating feature vector and each row vector in the learning parameter matrix to simulate the wandering pattern, and uses the distance topology between the biological heat map generating feature vector and each row vector in the learning parameter matrix as a wandering map. In this way, the wandering result of the wandering node in a specific wandering pattern on the wandering map is used to obtain the biological heat map entropy degradation maximization simulation value. Furthermore, based on the comparison between each biological heat map entropy degradation maximization simulation value in the set of biological heat map entropy degradation maximization simulation values ​​and a preset threshold, the part of the learning parameter matrix that does not contribute significantly to the input data representation and may even introduce noise or redundant information is screened out. Specifically, based on the comparison result, the learning parameter matrix is ​​gated and pruned to obtain the biological heat map learning parameter entropy degradation optimization matrix. Finally, the structure-optimized learning parameter matrix is ​​used as a feature modulation unit to linearly modulate the biological heat map generation feature vector to obtain the sparse optimized biological heat map generation feature vector.

[0052] Furthermore, the sparse optimized bioheatmap generated feature vector contains a variety of important information related to biodiversity, including species distribution, environmental factors and their interrelationships. These features provide a comprehensive perspective that reflects the complexity and diversity of ecosystems. Through the generator, the sparse optimized bioheatmap generated feature vector is converted into a heat map, which can make these abstract data more visual and make complex ecological information clear at a glance. By visualizing the bioheatmap, researchers can more easily identify the abundance of species and their distribution in different regions. This information is crucial for ecological monitoring and conservation decision-making, and can help identify high-value conservation areas and guide resource allocation and the formulation of management strategies.

[0053] In another embodiment of the present application, firstly, the dynamic behavior of a multi-species random competition system under random perturbations is analyzed, the existence conditions of a unique positive solution and its random persistence under the influence of white noise are explored, and the Has'minskii theory is used to study the invariant distribution and extinction, and the Milstein method is used for numerical simulation verification; secondly, the parameters in the multi-species food chain model without the main term are studied to affect the system dynamics under random perturbations, and the V function is constructed using the Lyapunov theorem to prove the uniqueness of the positive solution and the asymptotic stability of each equilibrium point; finally, the random perturbation behavior of the Holling II type food chain model without the main term is investigated, and its equilibrium point and stability are analyzed in detail, and the research results are verified by numerical simulation. Among them, for the problem of the dynamic behavior of a multi-species random competition system under random perturbations, the technical point is to use the application skills of inequalities and comparison principles to prove the existence and uniqueness of the positive solution, construct a new stochastic differential equation and apply the comparison principle to study the persistence and non-persistence in the sense of time mean. For the dynamic behavior of parameters in the multi-species food chain model without principal terms under random perturbations, the technical point is to construct a V function to prove whether the positive solution of the random system is unique, and to discuss the asymptotic stability of the random system at each equilibrium point of the deterministic system. For the dynamic behavior of the Holling II type multi-species food chain model without principal terms under random perturbations, the technical point is to discuss the asymptotic behavior of the random system at each equilibrium point of the deterministic system, and to verify the conclusions through numerical simulation.

[0054] In another embodiment of the present application, it is intended to deeply study the dynamic behavior of a multi-species food chain model under random disturbances, including time-delay-free and time-delay systems, and the Holling II type functional response model. By applying the LV theorem and the Lyapunov analysis method, the existence and uniqueness of positive solutions of these systems, the boundedness of the P-order moment, the persistence and extinction in the mean sense will be explored. In addition, numerical simulation will be used to verify the theoretical results. On the technical level, first, the boundedness of the P-order moment of the positive solution is proved by cleverly using inequalities and comparison principles; secondly, a new stochastic differential equation is constructed and the comparison principle is applied to study the persistence and non-persistence of the system; finally, the existence and uniqueness of the positive solution of the random system and the stationary distribution are proved by using the LV function and the Hasminskii theory.

[0055] In summary, the embodiment of the present application first obtains the surface cover image collected by high-resolution remote sensing images, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally generates a biodiversity heat map through a generator, thereby improving the efficiency and accuracy of biodiversity monitoring and helping to understand the health of the ecosystem.

[0056] As described above, the data processing system 100 for biodiversity monitoring according to the embodiment of the present application can be implemented in various terminal devices. In one example, the data processing system 100 for biodiversity monitoring can be integrated into the terminal device as a software module and / or a hardware module. For example, the data processing system 100 for biodiversity monitoring can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the data processing system 100 for biodiversity monitoring can also be one of the many hardware modules of the terminal device.

[0057] Alternatively, in another example, the data processing system 100 for biodiversity monitoring and the terminal device may also be separate devices, and the data processing system 100 for biodiversity monitoring may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0058] Figure 5 FIG. 1 is a flow chart of a data processing method for biodiversity monitoring according to an embodiment of the present application. Figure 5 As shown, according to the data processing method for biodiversity monitoring according to the embodiment of the present application, it includes: S110, acquiring a surface cover image collected by a high-resolution remote sensing image, environmental sound information collected by a sound sensor, and biological survey data collected by a database; S120, extracting a biological multimodal information association feature vector and a biological survey data global feature vector from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database; S130, generating a biodiversity heat map based on the biological multimodal information association feature vector and the biological survey data global feature vector.

[0059] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned data processing method for biodiversity monitoring have been described in detail above. Figures 1 to 4 The data processing system for biodiversity monitoring has been described in detail in the description of the data processing system for biodiversity monitoring, and therefore, its repeated description will be omitted.

[0060] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0061] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0062] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0063] Those skilled in the art can understand that all or part of the steps of the above method embodiments can be completed by hardware related to program instructions, and the above program can be stored in a readable storage medium. When the program is executed, the steps of the above method embodiments are executed; and the above storage medium includes: Various media that can store program codes, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks.

[0064] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

Claims

1. A data processing system for biodiversity monitoring, characterized in that: include: Biodiversity monitoring data acquisition module, used to obtain land cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases; A biodiversity monitoring data processing module, used to extract biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database; The biodiversity heat map generation module is used to generate a biodiversity heat map based on the biological multimodal information association feature vector and the biological survey data global feature vector.

2. The data processing system for biodiversity monitoring according to claim 1, characterized in that: The biodiversity monitoring data processing module comprises: A land cover image feature extraction unit, used for extracting features from the land cover image collected by the high-resolution remote sensing image to obtain a land cover feature vector of interest; An environmental sound information feature extraction unit, used to extract features from the environmental sound information collected by the sound sensor to obtain a global feature vector of the environmental sound audio; A biological multimodal information feature association unit, used for associating the ground cover feature vector of interest with the environmental sound audio global feature vector to obtain the biological multimodal information association feature vector; The biological survey data feature extraction unit is used to extract features from the biological survey data collected from the database to obtain a global feature vector of the biological survey data.

3. The data processing system for biodiversity monitoring according to claim 2, characterized in that: The ground cover image feature extraction unit comprises: A ground cover image target detection subunit is used to perform target detection on the ground cover image collected by the high-resolution remote sensing image to obtain a ground cover area of ​​interest; A land cover significant convolutional coding subunit, used for passing the land cover region of interest through a land cover significant object detector of interest to obtain a land cover feature map of interest; The land cover feature of interest pooling subunit is used to perform average value pooling on the land cover feature of interest map based on a feature matrix with a preset size to obtain a land cover feature of interest vector.

4. The data processing system for biodiversity monitoring according to claim 3, characterized in that: The environmental sound information feature extraction unit comprises: Extracting Mel-frequency cepstral coefficients from the ambient sound information collected by the sound sensor to obtain a plurality of short-time eigenvalues ​​of Mel-frequency cepstral coefficients of the ambient sound information; The multiple short-time eigenvalues ​​of the ambient sound information Mel-frequency cepstral coefficients are passed through an ambient sound information Mel-frequency cepstral feature encoder based on a convolutional neural network model to obtain the ambient sound audio global feature vector.

5. The data processing system for biodiversity monitoring according to claim 4, characterized in that: The biological multimodal information feature association unit comprises: Add dependency libraries through Maven; Use DruidDataSource to configure the data source, including database connection information, connection pool size, and connection timeout; When the application starts, initialize DruidDataSource and set it as a globally accessible data source; Obtaining the interesting feature vector of ground coverage and the global feature vector of ambient sound audio; In calculateMultiModalFeatureVector, two feature vectors are received as input, the product of the two feature vectors is calculated, the biological multimodal information association feature vector is obtained, and the calculation result is returned; The calculated biological multimodal information associated feature vector is stored back in the database.

6. The data processing system for biodiversity monitoring according to claim 5, characterized in that: The biological survey data feature extraction unit comprises: The biological survey data collected from the database is passed through a biological survey data text semantic coding model to obtain a plurality of biological survey data feature vectors; The plurality of biological survey data feature vectors are concatenated into the biological survey data global feature vector.

7. The data processing system for biodiversity monitoring according to claim 6, characterized in that: The biodiversity heat map generation module comprises: A biodiversity data feature fusion unit, used to fuse the biological multimodal information association feature vector and the biological survey data global feature vector to obtain a biological heat map generation feature vector; A biodiversity data feature optimization unit, used for adjusting the bioheat map generation feature vector by maximizing entropy degradation based on the confusion of the learning parameter's own structural information to obtain a sparse optimized bioheat map generation feature vector; The biodiversity data feature generation unit is used to pass the sparse optimized biological heat map generation feature vector through a generator to generate a biodiversity heat map.

8. The data processing system for biodiversity monitoring according to claim 7, characterized in that: The biodiversity data feature optimization unit comprises: Extracting each row vector of the learning parameter matrix in the generator as a set of biological heat map learning parameter row vectors; Calculating the biological heat map entropy degradation maximization simulation value between the biological heat map generation feature vector and each biological heat map learning parameter row vector in the set of biological heat map learning parameter row vectors to obtain a set of biological heat map entropy degradation maximization simulation values; Based on the comparison between each biological heat map entropy degradation maximization simulation value in the set of biological heat map entropy degradation maximization simulation values ​​and a preset threshold, the learning parameter matrix is ​​gated and pruned to obtain a biological heat map learning parameter entropy degradation optimization matrix; The biological heat map generation feature vector is matrix multiplied with the biological heat map learning parameter entropy degradation optimization matrix to obtain the sparse optimized biological heat map generation feature vector.

9. A data processing method for biodiversity monitoring, characterized in that: include: Obtain land cover images collected by high-resolution remote sensing images, environmental sound information collected by sound sensors, and biological survey data collected by databases; Extracting biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database; A biodiversity heat map is generated based on the biological multimodal information association feature vector and the biological survey data global feature vector.

10. The data processing method for biodiversity monitoring according to claim 9, characterized in that: Extracting biological multimodal information association feature vectors and biological survey data global feature vectors from the surface cover image collected by the high-resolution remote sensing image, the environmental sound information collected by the sound sensor, and the biological survey data collected by the database, including: Performing feature extraction on the land cover image collected by the high-resolution remote sensing image to obtain a land cover feature vector of interest; Performing feature extraction on the ambient sound information collected by the sound sensor to obtain a global feature vector of the ambient sound audio; Associating the ground cover feature vector of interest with the ambient sound audio global feature vector to obtain the bio-multimodal information associated feature vector; Feature extraction is performed on the biological survey data collected from the database to obtain a global feature vector of the biological survey data.

Images