A biodiversity survey data integration method and system based on GIS technology

Abstract

The present invention discloses a method and system for integrating biodiversity survey data based on GIS technology, which relates to the technical field of data integration, including: obtaining geographic spatial data and vegetation distribution images of a survey area through GIS, segmenting the survey area into similar areas, quantifying terrain data, climate data, and soil data of the survey area according to the segmented similar areas, performing data classification aggregation on the data quantification results to form a multi-source data integration body, performing image preprocessing on the vegetation distribution image, constructing a student network to extract image features from image filtering data, constructing a deep confidence network to perform network matching on the multi-source data integration body and the vegetation features of the survey area according to the average weight, adjusting the weight of the multi-source data integration body, performing multiple network matching on the multi-source data integration body and the vegetation features of the survey area, generating biodiversity trend data, and realizing the integration of multi-source biodiversity survey data.

Description

Technical Field

[0001] The present invention relates to the technical field of data integration, and in particular to a method and system for integrating biodiversity survey data based on GIS technology. Background Art

[0002] Biodiversity is of great significance to maintaining the ecological stability of the earth. With the advancement of human science and technology, biodiversity conservation has gradually attracted attention. In this process, biodiversity data integration is of great value in promoting biodiversity conservation. At present, the integration of biodiversity data mostly adopts manual statistics or probability statistics in the set survey area. This method has the following problems:

[0003] (1) When manually counting the number of organisms, it is difficult to go deep into various terrains to conduct statistics due to the complex terrain conditions, which easily leads to omissions. Manual statistics are inefficient and are only suitable for counting the number of organisms in a small area.

[0004] (2) In probability statistics, the statistical workload can be reduced through probability calculations, and the biological population statistics of a smaller area can be used to estimate the biological population of a larger area. However, this method will produce different statistical results depending on the choice of probability function, and there will be large errors. Summary of the invention

[0005] The purpose of the present invention is to provide a biodiversity survey data integration method and system based on GIS technology to solve the problems of easy omissions, low efficiency, unsuitability for larger areas and large errors in the prior art.

[0006] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0007] The first aspect of the present invention provides a method for integrating biodiversity survey data based on GIS technology, comprising the following steps:

[0008] Obtain the geospatial data and vegetation distribution images of the survey area through the Geographic Information System (GIS), and divide the survey area into similar areas according to the latitude, longitude and altitude distribution;

[0009] According to the segmented similar areas, the terrain data, climate data and soil data of the survey area are quantified, and the data quantification results are aggregated into data categories to form a multi-source data integration body;

[0010] Performing image preprocessing on the vegetation distribution image to obtain image filtering data, constructing a student network to extract image features from the image filtering data using knowledge distillation;

[0011] Constructing a deep belief network, performing network matching on the multi-source data integration and vegetation characteristics of the survey area according to average weights, and generating biodiversity survey data;

[0012] The weight of the multi-source data integration body is adjusted, and multiple network matching is performed on the multi-source data integration body and the vegetation characteristics of the survey area to generate biodiversity trend data.

[0013] As a preferred solution of the present invention, the geographic spatial data and vegetation distribution images of the survey area are obtained through the geographic information system GIS, including:

[0014] The latitude and longitude of the center point of the survey area, the shape of the area, the area and the time tangent point are obtained, and based on the latitude and longitude of the center point, the shape of the area, the area and the time tangent point, the GIS system is requested to obtain the geographic spatial data and vegetation distribution image of the survey area.

[0015] As a preferred solution of the present invention, the survey area is segmented into similar areas according to the latitude, longitude and altitude distribution, including:

[0016] The latitude and longitude map of the survey area is evenly divided into 24×24 grid maps, and the average altitude within the grid is calculated;

[0017] Using an altitude propagation algorithm to segment the grid map into similar areas;

[0018] Firstly, any grid in the grid map is selected as the starting point of the spread, and the absolute difference between the starting point of the spread and the average altitude of all the grids adjacent to it is calculated;

[0019] If the absolute difference of the average altitude is less than 20 meters, the absolute difference of the altitudes of the edges of adjacent grids is calculated;

[0020] If the absolute difference in the altitudes of the edges of the adjacent grids is less than 20 meters, it is considered that the altitude change between the two grids is gentle and they are the same type of grids;

[0021] If the absolute difference in the altitudes of the edges of the adjacent grids is greater than 20 meters, it is considered that the altitudes between the two grids change dramatically and are not of the same type;

[0022] If the absolute difference between the average altitudes is greater than 20 meters, it is considered that the altitudes between the two grids change dramatically and are not of the same type;

[0023] A region dividing line is inserted between the non-similar grids, and the same grid is selected as a new propagation starting point. The above operation is repeated to traverse all grids until the region dividing line is closed to form a similar region segmentation result.

[0024] As a preferred solution of the present invention, data quantification is performed on the topographic data, climate data and soil data of the survey area according to the segmented similar areas, including:

[0025] Calculate the mean of each data within the same region;

[0026] Construct a data group for terrain data, and its expression is:

[0027] ;

[0028] in, Indicates the terrain type. Indicates the starting altitude of the terrain. Indicates the terrain end altitude, Indicates terrain area;

[0029] The annual average temperature value is calculated for the climate data, and the expression is:

[0030] ;

[0031] in, represents the annual average temperature value, Indicates the monthly average temperature value;

[0032] Construct a data set for soil data, and its expression is:

[0033] ;

[0034] in, Indicates soil pH. Indicates soil hardness.

[0035] As a preferred solution of the present invention, data quantification results are aggregated into data categories to form a multi-source data integration body, including:

[0036] The K-nearest neighbor algorithm is used to construct a sample network using the segmented similar regions as data samples, with the value of K being 10. The quantized data Te, T, and So are input into the sample network as knowledge samples, and the knowledge samples are normalized as node vectors of the network.

[0037] The data sample and the knowledge sample are first-order convolved, and the convolution result is passed to the next node along the topological structure of the sample network. The next node receives the convolution result and performs first-order convolution with the node vector and updates the node vector.

[0038] Cluster analysis is performed based on the output of sample network nodes to obtain a multi-source data integration body.

[0039] As a preferred solution of the present invention, image preprocessing is performed on the vegetation distribution image to obtain image filtering data, including:

[0040] Digitalizing the vegetation distribution image of the survey area to form an image gray value matrix, and performing dynamic median filtering on the image gray value matrix;

[0041] Using an observation window with a length of 3 and a width of 1 to observe the image gray value matrix, sorting the image gray values ​​within the field of view of the observation window, and assigning the median value of the sorting result to all matrix elements within the field of view of the observation window;

[0042] Move the observation window by the distance of two matrix elements, set the observation window length to 5, perform sorting again, and assign the median value in the sorting result to all matrix elements within the field of view of the observation window;

[0043] The observation window is moved by the distance of four matrix elements, and the length of the observation window is set to 3. The above operation is repeated to obtain the image filtering data. The filter function expression is:

[0044] ;

[0045] in, express OK Grayscale values ​​of matrix elements of columns, The value of is 2 or 4, Represents the median grayscale value of the image within the field of view of the observation window.

[0046] As a preferred solution of the present invention, a student network is constructed to extract image features from the image filtering data using knowledge distillation, including:

[0047] The teacher network is used to extract features from the image filtering data, and the vegetation types, quantities and distribution in the image are used to form a training data set. The feature vector output by the middle layer of the teacher network is used to supervise the student network training. The training data set is used as the training label for the student network training input. The overall objective function expression of the student model is:

[0048] ;

[0049] in, represents the weight value of the cross entropy loss, represents the loss function of the teacher network, represents the divergence of the student network prediction results;

[0050] The trained student network is used to perform feature extraction on the image filtering data.

[0051] As a preferred solution of the present invention, a deep belief network is constructed to perform network matching on the multi-source data integration body and the vegetation characteristics of the survey area according to the average weight to generate biodiversity survey data, including:

[0052] Randomly select 50% of the multi-source data integration, set the data weights of each part of it to be equal, and manually match them with vegetation features as pre-training materials for the deep belief network;

[0053] Inputting the pre-trained material into the interface layer of the deep belief network, each pre-trained material corresponds to an independent deep belief network neuron;

[0054] The initial restricted Boltzmann machine is pre-trained using the CD-7 algorithm, and the activation state of the lower-layer neurons is updated through the state transition probability function between the interface layer and the lower-layer neurons;

[0055] When the initial restricted Boltzmann machine parameter weights reach the optimal solution, the parameters are fixed and the lower restricted Boltzmann machine is repeatedly trained until all restricted Boltzmann machine parameter weights reach the optimal solution;

[0056] The deep belief network with the optimal solution of all restricted Boltzmann machine parameter weights is expanded into a BP neural network, and the artificial matching results are used as the network training output expectations. The restricted Boltzmann machine weights are fine-tuned until the network training output meets the artificial matching results. The error indicator expression between the network training output and the output expectation is:

[0057]

[0058] in, Represents the output result of network training, Indicates the output expectation, Indicates the number of training samples;

[0059] After the weight fine-tuning is completed, the remaining 50% of the multi-source data integration and the vegetation characteristics of the survey area are input into the deep belief network to obtain the network matching results and generate biodiversity survey data.

[0060] As a preferred solution of the present invention, the weight of the multi-source data integration body is adjusted, and multiple network matching is performed on the multi-source data integration body and the vegetation characteristics of the survey area to generate biodiversity trend data, including:

[0061] By adjusting the data weights of various parts within the multi-source data integration body for multiple times, re-inputting the data into the deep belief network for training;

[0062] In each training, one of the terrain, climate or soil is selected as the main influencing factor, and its weight is set to 70%, and the weights of the other two parts are set to 15%;

[0063] Different network matching results are obtained through multiple trainings. The different network matching results are compared with generalized biodiversity survey data to obtain the effects of different environmental factors on biodiversity.

[0064] The second aspect of the present invention is a biodiversity survey data integration system based on GIS technology, characterized by comprising:

[0065] GIS data request module: sends data acquisition commands to the GIS system according to user needs and receives returned data;

[0066] Image processing module: image sampling, segmentation and feature extraction of image data sent back by the GIS system;

[0067] Neural network module: trains the student model or deep belief network according to different training samples and training methods, and processes the new input samples according to the training parameters after the training.

[0068] Data integration platform: manages data transmission between modules, parses user commands, displays module information, and outputs data integration or processing results.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] The present invention adopts GIS technology, obtains the geographic spatial data and vegetation distribution images of the survey area through the GIS system, uses deep neural networks to process the data, and calculates the number of organisms in the survey area. It can complete biodiversity surveys over a very large range. After training and parameter adjustment, the deep neural network can accurately calculate the number of organisms without human intervention, thereby improving the efficiency of biodiversity surveys.

[0071] By changing the weights of different environmental factors in network matching and performing multiple matches, we can obtain the impact of different environmental factors on biodiversity and reduce the possible errors in a single match. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0073] Figure 1 A flow chart of a biodiversity survey data integration method based on GIS technology provided by an embodiment of the present invention;

[0074] Figure 2 A block diagram of a biodiversity survey data integration system based on GIS technology provided by an embodiment of the present invention;

[0075] 1-GIS data request module; 2-image processing module; 3-neural network module; 4-data integration platform. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] like Figure 1 As shown, the present invention provides a method for integrating biodiversity survey data based on GIS technology, comprising the following steps:

[0078] The geographic information system (GIS) is used to obtain the geospatial data and vegetation distribution images of the survey area, and the survey area is divided into similar areas according to the latitude, longitude and altitude distribution;

[0079] In this embodiment, the geographic information system GIS contains geographic information on the earth's surface and atmospheric space, including location information, historical climate data, terrain features and satellite images. By obtaining the original information through the GIS system and analyzing it, accurate data on the earth's surface and atmospheric space can be obtained.

[0080] In this embodiment, since the terrain conditions within the survey area are complex and varied, similar area segmentation divides the areas with similar terrain features within the survey area according to terrain transformation, and regards similar areas as the same entity, thereby reducing the workload of data analysis.

[0081] According to the segmented similar areas, the terrain data, climate data, and soil data of the survey area are quantified, and the data quantification results are aggregated to form a multi-source data integration body;

[0082] In this embodiment, the terrain data, climate data and soil data in the GIS system are numerous and huge in number. Data quantification selects the most representative indicators in the data and quantifies the indicators into matrix form or numerical form, which is beneficial to the subsequent data processing of the system.

[0083] In this embodiment, terrain data, climate data and soil data are multi-source descriptive data for the same survey area. Data class aggregation aggregates the multi-source data of the same survey area into the same type, integrates the multi-source data, and enhances the accuracy of the description of the environmental factors of the survey area.

[0084] Perform image preprocessing on vegetation distribution images, obtain image filtering data, build a student network and use knowledge distillation to extract image features from image filtering data;

[0085] In this embodiment, the image preprocessing adopts a dynamic window to perform median filtering. Median filtering is a nonlinear digital filtering method that has the ability to maximize the retention of edge image features and has a good ability to remove image noise.

[0086] Construct a deep belief network to perform network matching on the multi-source data integration and vegetation characteristics of the survey area according to the average weight to generate biodiversity survey data;

[0087] In this embodiment, the deep belief network trains itself by learning the information in the pre-training materials, and expresses the characteristic information of the training results through the connection weights between neurons. It uses the method of step-by-step training of the restricted Boltzmann machine and fine-tuning of the global parameters to avoid falling into the problem of local optimal solutions.

[0088] Adjust the weight of the multi-source data integration body, perform multiple network matching on the multi-source data integration body and the vegetation characteristics of the survey area, and generate biodiversity trend data;

[0089] In this embodiment, the weights of different parts within the multi-source data integration body are changed. When the weights are different, the matching results of the multi-source data integration body and the vegetation characteristics of the survey area will also be different. By controlling the variables and comparing the matching results under different weights, the correlation between the vegetation characteristics and the different parts within the multi-source data integration body can be analyzed.

[0090] The geographic information system (GIS) is used to obtain geospatial data and vegetation distribution images of the survey area, including:

[0091] Determine the longitude and latitude of the center point of the survey area, the shape of the area, the area of ​​the area and the time tangent point, and request the GIS system to obtain the geospatial data and vegetation distribution images of the survey area based on the longitude and latitude of the center point, the shape of the area, the area of ​​the area and the time tangent point;

[0092] In this embodiment, the position of the survey area on the earth's surface is located by determining the longitude and latitude of the center point of the survey area, the expansion range of the survey area is limited by determining the shape and area of ​​the area, and the position of the survey area on the time axis is limited by determining the time node.

[0093] The survey area is divided into similar areas according to the latitude, longitude and altitude distribution, including:

[0094] The latitude and longitude map of the survey area is evenly divided into 24×24 grid maps, and the average elevation within the grid is calculated;

[0095] Use the elevation propagation algorithm to segment the grid map into similar areas;

[0096] First, select any grid in the grid map as the starting point of the spread, and calculate the absolute difference between the starting point of the spread and the average elevation of all adjacent grids;

[0097] If the absolute difference in the average altitude is less than 20 meters, then calculate the absolute difference in the altitudes of the adjacent grid edges;

[0098] If the absolute difference in the elevation of the edges of adjacent grids is less than 20 meters, it is considered that the elevation change between the two grids is gentle and they are of the same type;

[0099] If the absolute difference in the elevation of the edges of adjacent grids is greater than 20 meters, it is considered that the elevation change between the two grids is drastic and they are not of the same type;

[0100] If the absolute difference in the average elevation is greater than 20 meters, it is considered that the elevation change between the two grids is drastic and they are not of the same type;

[0101] Insert a region dividing line between non-similar grids, select a similar grid as a new spreading starting point, repeat the above operation to traverse all grids until the region dividing line is closed, and form a similar region segmentation result;

[0102] In this embodiment, the altitude propagation algorithm is sensitive to the altitude parameters of the survey area. If the altitude inside the survey area undergoes a large degree of mutation, it is considered that the terrain of the area has changed dramatically. In the altitude propagation algorithm, the edge of the area is used as the edge of the terrain segmentation. By continuously iterating and traversing all grids, the terrain segmentation of the survey area can be completed. In the absence of other auxiliary conditions or prior knowledge, the altitude propagation algorithm can achieve the best segmentation performance in complex terrain.

[0103] According to the segmented similar areas, the terrain data, climate data, and soil data of the survey area are quantified, including:

[0104] Calculate the mean of each data within the same region;

[0105] Construct a data group for terrain data, and its expression is:

[0106] ;

[0107] in, Indicates the terrain type. Indicates the starting altitude of the terrain. Indicates the terrain end altitude, Indicates terrain area;

[0108] The annual average temperature value is calculated for the climate data, and the expression is:

[0109] ;

[0110] in, represents the annual average temperature value, Indicates the monthly average temperature value;

[0111] Construct a data set for soil data, and its expression is:

[0112]

[0113] in, Indicates soil pH. Indicates soil hardness;

[0114] In this embodiment, the terrain category data is text data, and various terrains are marked with corresponding fixed data. The starting altitude, ending altitude, and terrain area are numerical data, and the values ​​can be increased or decreased during the program running process. The annual average temperature is calculated by averaging the monthly average temperature to avoid the workload of calculating the daily average temperature.

[0115] Data quantification results are aggregated into data categories to form a multi-source data integration, including:

[0116] The K-nearest neighbor algorithm is used to construct a sample network using the segmented similar regions as data samples. The value of K is 10. The quantized data Te, T, and So are input into the sample network as knowledge samples. The knowledge samples are normalized and used as node vectors of the network.

[0117] The data sample and the knowledge sample are first-order convolved, and the convolution result is passed to the next node along the topological structure of the sample network. The next node receives the convolution result and performs first-order convolution with the node vector and updates the node vector.

[0118] Perform cluster analysis based on sample network node output to obtain multi-source data integration;

[0119] In this embodiment, the K-nearest neighbor algorithm classifies the input data into the most frequent category of the sample category by finding the K nearest samples around the sample point and determining the sample category. The K-nearest neighbor method is used to construct a sample network. There is no need to train the sample network in advance. Sample clustering can be performed directly using existing knowledge samples, thereby reducing the time complexity of training.

[0120] In this embodiment, hierarchical clustering of samples is achieved through convolution and vector updating. The clustering results of the first level are used as the input of the second level. The existing knowledge samples are mixed with the clustering results of the first level as knowledge samples for clustering at the second level. This can increase the number of knowledge samples and improve clustering accuracy.

[0121] Perform image preprocessing on vegetation distribution images to obtain image filtering data, including:

[0122] The vegetation distribution images of the survey area are digitized to form an image gray value matrix, and the image gray value matrix is ​​subjected to dynamic median filtering;

[0123] An observation window with a length of 3 and a width of 1 is used to observe the image gray value matrix, the image gray values ​​within the field of view of the observation window are sorted, and the median value of the sorting result is assigned to all matrix elements within the field of view of the observation window;

[0124] Move the observation window by the distance of two matrix elements, set the observation window length to 5, sort again, and assign the median value of the sorting result to all matrix elements within the observation window field of view;

[0125] Move the observation window by the distance of four matrix elements and set the observation window length to 3. Repeat the above operation to obtain the image filtering data. The filter function expression is:

[0126]

[0127] in, express OK Grayscale values ​​of matrix elements of columns, The value of is 2 or 4, Represents the median grayscale value of the image within the field of view of the observation window;

[0128] In this embodiment, dynamic window filtering is used, and the window size is changed by considering the position of the matrix elements when selecting the window size, thereby avoiding data diffusion at the edge of the matrix and preventing the edge of the filtered image from ghosting or blurring.

[0129] Construct a student network to extract image features from image filtering data using knowledge distillation, including:

[0130] The teacher network is used to extract features from the image filtering data, and the vegetation types, quantities and distribution in the image are used to form a training data set. The feature vector output by the middle layer of the teacher network is used to supervise the student network training. The training data set is used as the training label for the student network training input. The overall objective function expression of the student model is:

[0131] ;

[0132] in, represents the weight value of the cross entropy loss, represents the loss function of the teacher network, represents the divergence of the student network prediction results;

[0133] Use the trained student network to extract features from image filtering data;

[0134] In this embodiment, the teacher network has a very large network scale and network complexity, and has relatively excellent performance in feature extraction, but the teacher network consumes a lot of computing resources, which is not conducive to network migration. By using the output results of the teacher network as training labels for the student network, a smaller-scale student network is used to imitate the neuron structure of the teacher network, and a student network with performance close to that of the teacher network is obtained with less computing resource consumption, thereby reducing the difficulty of network migration.

[0135] Construct a deep belief network, match the multi-source data integration and the vegetation characteristics of the survey area according to the average weight, and generate biodiversity survey data, including:

[0136] Randomly select 50% of the multi-source data integration, set the data weights of each part of it to be equal, and manually match them with vegetation features as pre-training materials for the deep belief network;

[0137] Input the pre-trained materials into the interface layer of the deep belief network, where each pre-trained material corresponds to an independent deep belief network neuron;

[0138] The initial restricted Boltzmann machine is pre-trained using the CD-7 algorithm, and the activation state of the lower-layer neurons is updated through the state transition probability function between the interface layer and the lower-layer neurons;

[0139] When the initial restricted Boltzmann machine parameter weights reach the optimal solution, fix the parameters and repeatedly train the lower restricted Boltzmann machine until all restricted Boltzmann machine parameter weights reach the optimal solution;

[0140] The deep belief network with the optimal solution of all restricted Boltzmann machine parameter weights is expanded into a BP neural network, and the artificial matching results are used as the network training output expectations. The restricted Boltzmann machine weights are fine-tuned until the network training output meets the artificial matching results. The error indicator expression between the network training output and the output expectation is:

[0141]

[0142] in, Represents the output result of network training, Indicates the output expectation, Indicates the number of training samples;

[0143] After the weight fine-tuning is completed, the remaining 50% of the multi-source data integration and the vegetation characteristics of the survey area are input into the deep belief network to obtain the network matching results and generate biodiversity survey data;

[0144] In this embodiment, the basic structure of the deep belief network is a restricted Boltzmann machine, and a complete deep belief network is constructed by the step-by-step connection of restricted Boltzmann machines. When the deep belief network is trained, the first layer of restricted Boltzmann machines is trained first. When the parameters are fixed, the lower layer of restricted Boltzmann machines are trained until the pre-training of the deep belief network is completed. This process is an unsupervised training mode and does not require the input of a large number of samples for manual matching results.

[0145] In this embodiment, after the pre-training is completed, the parameters of the restricted Boltzmann machine are fine-tuned through the back propagation network. At this time, the manual matching results of the samples are used as supervised data for supervised training, ensuring that the final output results of the deep belief network are maximized to fit the manual matching results.

[0146] Adjust the weight of the multi-source data integration, perform multiple network matching on the multi-source data integration and vegetation characteristics of the survey area, and generate biodiversity trend data, including:

[0147] Change the data weights of each part of the multi-source data integration body multiple times and re-input it into the deep belief network for training;

[0148] In each training, one of the terrain, climate or soil is selected as the main influencing factor, and its weight is set to 70%, and the weights of the other two parts are set to 15%;

[0149] Through multiple trainings, different network matching results are obtained, and the different network matching results are compared with generalized biodiversity survey data to obtain the effects of different environmental factors on biodiversity;

[0150] In this embodiment, the matching results of the multi-source data integration with different weight allocations and the vegetation characteristics are different. The main influencing factor causing the difference is the part with the largest weight in the weight allocation. Through multiple weight allocations and deep belief network matching, the main environmental influencing factors of different vegetation characteristics can be found.

[0151] like Figure 2 As shown, the present invention provides a biodiversity survey data integration system based on GIS technology, comprising:

[0152] GIS data request module, image processing module, neural network module, data integration platform;

[0153] The GIS data request module sends data acquisition commands to the GIS system according to user needs and receives the returned data;

[0154] The image processing module performs image sampling, segmentation and feature extraction on the image data sent back by the GIS system;

[0155] The neural network module trains the student model or deep belief network according to different training samples and training methods, and processes the new input samples according to the training parameters after the training.

[0156] The data integration platform manages data transmission between modules, parses user commands, displays module information, and outputs data integration or processing results;

[0157] In this embodiment, the data integration platform is the top-level module of the biodiversity survey data integration system, which is responsible for managing data transmission between modules, parsing user commands, displaying module information, and outputting data integration or processing results. The GIS data request module, under the control of the data integration platform, issues data acquisition commands to the GIS system according to user commands and receives return data. The image processing module, under the control of the data integration platform, performs image sampling, segmentation and feature extraction on the return data received from the GIS data request module. The neural network module, under the control of the data integration platform, trains the student model or the deep belief network according to different training samples and training methods. After the training, the new input samples are processed according to the training parameters and the processing results are transmitted to the data integration platform for display.

[0158] The present invention adopts GIS technology, obtains the geographic spatial data and vegetation distribution images of the survey area through the GIS system, uses deep neural networks to process the data, and calculates the number of organisms in the survey area. It can complete biodiversity surveys over a very large range. After training and parameter adjustment, the deep neural network can accurately calculate the number of organisms without human intervention, thereby improving the efficiency of biodiversity surveys.

[0159] By changing the weights of different environmental factors in network matching and performing multiple matches, we can obtain the impact of different environmental factors on biodiversity and reduce the possible errors in a single match.

[0160] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A biodiversity survey data integration method based on GIS technology, characterized in that: The following steps are involved: Obtain the geospatial data and vegetation distribution images of the survey area through the Geographic Information System (GIS), and divide the survey area into similar areas according to the latitude, longitude and altitude distribution; According to the segmented similar areas, the terrain data, climate data and soil data of the survey area are quantified, and the data quantification results are aggregated into data categories to form a multi-source data integration body; Performing image preprocessing on the vegetation distribution image to obtain image filtering data, constructing a student network to extract image features from the image filtering data using knowledge distillation; Constructing a deep belief network, performing network matching on the multi-source data integration and vegetation characteristics of the survey area according to average weights, and generating biodiversity survey data; Adjusting the weight of the multi-source data integration body, performing multiple network matching on the multi-source data integration body and vegetation characteristics of the survey area, and generating biodiversity trend data; Construct a deep belief network, perform network matching on the multi-source data integration and vegetation characteristics of the survey area according to the average weight, and generate biodiversity survey data, including: Randomly select 50% of the multi-source data integration, set the data weights of each part of it to be equal, and manually match them with vegetation features as pre-training materials for the deep belief network; Inputting the pre-trained material into the interface layer of the deep belief network, each pre-trained material corresponds to an independent deep belief network neuron; The initial restricted Boltzmann machine is pre-trained using the CD-7 algorithm, and the activation state of the lower-layer neurons is updated through the state transition probability function between the interface layer and the lower-layer neurons; When the initial restricted Boltzmann machine parameter weights reach the optimal solution, the parameters are fixed and the lower restricted Boltzmann machine is repeatedly trained until all restricted Boltzmann machine parameter weights reach the optimal solution; The deep belief network with the optimal solution of all restricted Boltzmann machine parameter weights is expanded into a BP neural network, and the artificial matching results are used as the network training output expectations. The restricted Boltzmann machine weights are fine-tuned until the network training output meets the artificial matching results. The error indicator expression between the network training output and the output expectation is: in, Represents the output result of network training, Indicates the output expectation, Indicates the number of training samples; After the weight fine-tuning is completed, the remaining 50% of the multi-source data integration and the vegetation characteristics of the survey area are input into the deep belief network to obtain the network matching results and generate biodiversity survey data; The weight of the multi-source data integration body is adjusted, and multiple network matching is performed on the multi-source data integration body and the vegetation characteristics of the survey area to generate biodiversity trend data, including: By adjusting the data weights of various parts within the multi-source data integration body for multiple times, re-inputting the data into the deep belief network for training; In each training, one of the terrain, climate or soil is selected as the main influencing factor, and its weight is set to 70%, and the weights of the other two parts are set to 15%; Different network matching results are obtained through multiple trainings. The different network matching results are compared with generalized biodiversity survey data to obtain the effects of different environmental factors on biodiversity.

2. The method for integrating biodiversity survey data based on GIS technology according to claim 1, characterized in that: The geographic information system (GIS) is used to obtain geospatial data and vegetation distribution images of the survey area, including: The latitude and longitude of the center point of the survey area, the shape of the area, the area and the time tangent point are obtained, and based on the latitude and longitude of the center point, the shape of the area, the area and the time tangent point, the GIS system is requested to obtain the geographic spatial data and vegetation distribution image of the survey area.

3. The method for integrating biodiversity survey data based on GIS technology according to claim 1, characterized in that: The survey area is divided into similar areas according to the latitude, longitude and altitude distribution, including: The latitude and longitude map of the survey area is evenly divided into 24×24 grid maps, and the average altitude within the grid is calculated; Using an altitude propagation algorithm to segment the grid map into similar areas; Firstly, any grid in the grid map is selected as the starting point of the spread, and the absolute difference between the starting point of the spread and the average altitude of all the grids adjacent to it is calculated; If the absolute difference of the average altitude is less than 20 meters, the absolute difference of the altitudes of the edges of adjacent grids is calculated; If the absolute difference in the altitudes of the edges of the adjacent grids is less than 20 meters, it is considered that the altitude change between the two grids is gentle and they are the same type of grids; If the absolute difference in the altitudes of the edges of the adjacent grids is greater than 20 meters, it is considered that the altitudes between the two grids change dramatically and are not of the same type; If the absolute difference between the average altitudes is greater than 20 meters, it is considered that the altitudes between the two grids change dramatically and are not of the same type; A region dividing line is inserted between the non-similar grids, and the same grid is selected as a new propagation starting point. The above operation is repeated to traverse all grids until the region dividing line is closed to form a similar region segmentation result.

4. The method for integrating biodiversity survey data based on GIS technology according to claim 3, characterized in that: According to the segmented similar areas, the terrain data, climate data, and soil data of the survey area are quantified, including: Calculate the mean of each data within the same region; Construct a data group for terrain data, and its expression is: ； in, Indicates the terrain type. Indicates the starting altitude of the terrain. Indicates the terrain end altitude, Indicates terrain area; The annual average temperature value is calculated for the climate data, and the expression is: ； in, represents the annual average temperature value, Indicates the monthly average temperature value; Construct a data set for soil data, and its expression is: ； in, Indicates soil pH. Indicates soil hardness.

5. The method for integrating biodiversity survey data based on GIS technology according to claim 4, characterized in that: Data quantification results are aggregated into data categories to form a multi-source data integration, including: The K-nearest neighbor algorithm is used to construct a sample network using the segmented similar regions as data samples, with the value of K being 10. , , Inputting the sample network as a knowledge sample, normalizing the knowledge sample and using it as a node vector of the network; The data sample and the knowledge sample are first-order convolved, and the convolution result is passed to the next node along the topological structure of the sample network. The next node receives the convolution result and performs first-order convolution with the node vector and updates the node vector. Cluster analysis is performed based on the output of sample network nodes to obtain a multi-source data integration body.

6. The method for integrating biodiversity survey data based on GIS technology according to claim 5, characterized in that: Performing image preprocessing on the vegetation distribution image to obtain image filtering data includes: Digitalizing the vegetation distribution image of the survey area to form an image gray value matrix, and performing dynamic median filtering on the image gray value matrix; Using an observation window with a length of 3 and a width of 1 to observe the image gray value matrix, sorting the image gray values ​​within the field of view of the observation window, and assigning the median value of the sorting result to all matrix elements within the field of view of the observation window; Move the observation window by the distance of two matrix elements, set the observation window length to 5, perform sorting again, and assign the median value in the sorting result to all matrix elements within the field of view of the observation window; The observation window is moved by the distance of four matrix elements, and the length of the observation window is set to 3. The above operation is repeated to obtain the image filtering data. The filter function expression is: ； in, express OK Grayscale values ​​of matrix elements of columns, The value of is 2 or 4, Represents the median grayscale value of the image within the field of view of the observation window.

7. The method for integrating biodiversity survey data based on GIS technology according to claim 6, characterized in that: Constructing a student network to extract image features from the image filtering data using knowledge distillation, including: The teacher network is used to extract features from the image filtering data, and the vegetation types, quantities and distribution in the image are used to form a training data set. The training data set is used as the training label for the student network training input. The feature vector output by the middle layer of the teacher network is used to supervise the student network training. The overall objective function expression of the student model is: ； in, represents the weight value of the cross entropy loss, represents the loss function of the teacher network, represents the divergence of the student network prediction results; The trained student network is used to perform feature extraction on the image filtering data.

8. A GIS-based biodiversity survey data integration system according to any one of claims 1 to 7, characterized in that: The system comprises: GIS data request module: sends data acquisition commands to the GIS system according to user needs and receives returned data; Image processing module: image sampling, segmentation and feature extraction of image data sent back by the GIS system; Neural network module: trains the student model or deep belief network according to different training samples and training methods, and processes the new input samples according to the training parameters after the training. Data integration platform: manages data transmission between modules, parses user commands, displays module information, and outputs data integration or processing results.

Images