Geographic Information Remote Sensing Mapping System and Method
By employing deep learning techniques and a bi-branch convolutional neural network model, the problems of different spectra for the same object and the same spectra for different objects in remote sensing image classification were solved, achieving accurate segmentation of remote sensing spectral images and identification and classification of ground objects, thus improving classification accuracy.
Patent Information
- Application Number
- CN202310312360.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-28
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2043-03-28
AI Technical Summary
Existing technologies for remote sensing image classification suffer from problems such as different spectra for the same object and the same spectra for different objects. This leads to issues such as mixing water bodies and mountain shadows, unclear boundaries between unused land and urban public transport land, misclassifying rivers as urban public transport land, confusing forest land and cultivated land, and misclassifying mountain shadows as urban public transport land, resulting in low classification accuracy.
Using deep learning-based artificial intelligence technology, through multispectral image acquisition, directional gradient histogram transformation, multi-channel aggregation, image segmentation, multi-scale feature extraction, and global feature association, a bi-branch convolutional neural network model is used to extract the implicit feature information of ground objects in remote sensing spectral images and perform image semantic segmentation.
It achieves accurate identification and classification of remote sensing spectral images, accurately segmenting images into sub-regions of land cover categories with different pixel properties, thereby improving classification accuracy.
Smart Images

Figure CN116452973B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent surveying and mapping technology, and more specifically, to a geographic information remote sensing surveying and mapping system and method thereof. Background Art
[0002] Remote sensing mapping mainly refers to mapping using electromagnetic wave signals reflected, scattered or emitted by objects received by sensors, and is often used to map the surface of the earth. Remote sensing images are films (or photos) that only record the size of electromagnetic waves of various objects. In remote sensing image maps, the content elements of the map are mainly composed of images, assisted by certain map symbols to express or explain the mapping objects. Compared with ordinary maps, image maps have rich ground information, clear content levels, and clear and easy-to-read maps, fully demonstrating the dual advantages of images and maps.
[0003] At present, when classifying remote sensing images, there are several major problems in supervised classification due to the existence of problems such as different spectra for the same object and the same spectra for different objects: the confusion of water bodies and mountain shadows; the unclear classification boundary between unused land and urban public transportation land; rivers may be misjudged as urban public transportation land; forest land and cultivated land are confused; mountain shadows may be misjudged as urban public transportation land.
[0004] Therefore, an optimized geographic information remote sensing mapping system 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 geographic information remote sensing mapping system and method thereof, which obtains a remote sensing multispectral image to be processed; uses artificial intelligence technology based on deep learning to mine the spectral features of the objects in the remote sensing spectral image, and extracts the implicit feature information of the objects in the remote sensing spectral image based on the spectral features of the objects, so as to classify all the pixels of the image into several categories according to their properties. In this way, the objects can be accurately identified and classified, so as to accurately segment the remote sensing spectral image into several object category sub-regions with different categories of pixel properties.
[0006] In a first aspect, a geographic information remote sensing mapping system is provided, comprising:
[0007] A multispectral image acquisition module is used to obtain remote sensing multispectral images to be processed;
[0008] A directional gradient histogram transformation module, used for calculating the directional gradient histogram of the remote sensing multispectral image to be processed;
[0009] A multi-channel aggregation module, used for aggregating the remote sensing multispectral image to be processed and the directional gradient histogram along the channel dimension to obtain a multi-channel image;
[0010] An image block module, used for performing image block processing on the multi-channel image to obtain a sequence of image blocks;
[0011] A multi-scale feature extraction module, used for respectively passing the sequence of image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales;
[0012] A global feature association module, used for arranging the sequence of image block feature matrices into a multi-channel global feature matrix; and
[0013] The semantic segmentation module is used to perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result.
[0014] In the above-mentioned geographic information remote sensing mapping system, the multi-scale feature extraction module includes: a first-scale image feature extraction unit, which is used to use the first convolutional neural network model with the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix; a second-scale image feature extraction unit, which is used to use the second convolutional neural network model with the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a second branch feature matrix; and a multi-scale feature fusion unit, which is used to fuse the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix.
[0015] In the above-mentioned geographic information remote sensing mapping system, the first-scale image feature extraction unit is used to: use each layer of the first convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer to output the first branch feature matrix from the last layer of the first convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the first convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
[0016] In the above-mentioned geographic information remote sensing mapping system, the second-scale image feature extraction unit is used to: use each layer of the second convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer to output the second branch feature matrix from the last layer of the second convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the second convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
[0017] In the above-mentioned geographic information remote sensing mapping system, the multi-scale feature fusion unit is used to: adopt convolutional dictionary control response learning to fuse the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix; wherein the optimization formula is:
[0018]
[0019] Where M1 and M2 are the first branch feature matrix and the second branch feature matrix respectively, and ‖·‖ F represents the Frobenius norm of the matrix, represents matrix addition, Represents matrix multiplication, M c represents the image block feature matrix, (·) T Represents the transpose of a matrix.
[0020] In the above-mentioned geographic information remote sensing mapping system, the global feature association module is used to: arrange the sequence of the image block feature matrix into a multi-channel global feature matrix according to the following arrangement formula; wherein the arrangement formula is:
[0021] M c =Concat[M1, M2…M n ]
[0022] Among them, M1, M2…M n represents the sequence of the image block feature matrix, Concat[·] represents the cascade function, M c represents the multi-channel global feature matrix.
[0023] In the above-mentioned geographic information remote sensing mapping system, the image semantic segmentation result is to segment the remote sensing multispectral image to be processed into several ground object category sub-regions with different categories of pixel properties.
[0024] In a second aspect, a geographic information remote sensing mapping method is provided, which includes:
[0025] Acquire remote sensing multispectral images to be processed;
[0026] Calculating a directional gradient histogram of the remote sensing multispectral image to be processed;
[0027] Aggregating the remote sensing multispectral image to be processed and the oriented gradient histogram along a channel dimension to obtain a multi-channel image;
[0028] Performing image block processing on the multi-channel image to obtain a sequence of image blocks;
[0029] Passing the sequence of image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales;
[0030] Arranging the sequence of image block feature matrices into a multi-channel global feature matrix; and
[0031] Perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result
[0032] In the above-mentioned geographic information remote sensing mapping method, the sequence of image blocks is respectively passed through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales, including: using the first convolutional neural network model with the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix; using the second convolutional neural network model with the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a second branch feature matrix; and, fusing the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix.
[0033] In the above-mentioned geographic information remote sensing mapping method, the first convolutional neural network model with a dual-branch structure is used to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix, including: using each layer of the first convolutional neural network model with a dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer to output the first branch feature matrix from the last layer of the first convolutional neural network model with a dual-branch structure, wherein the input of the first layer of the first convolutional neural network model with a dual-branch structure is each image block in the sequence of image blocks.
[0034] Compared with the prior art, the geographic information remote sensing mapping system and method provided by the present application obtains remote sensing multispectral images to be processed; uses artificial intelligence technology based on deep learning to mine the spectral features of objects in remote sensing spectral images, extracts implicit feature information of objects in remote sensing spectral images based on the spectral features of objects, and divides all pixels of the image into several categories according to their properties. In this way, objects can be accurately identified and classified, so as to accurately segment remote sensing spectral images into several object category sub-regions with different pixel properties. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0036] Figure 1 This is a diagram of an application scenario of a geographic information remote sensing mapping system according to an embodiment of the present application.
[0037] Figure 2 4 is a block diagram of a geographic information remote sensing mapping system according to an embodiment of the present application.
[0038] Figure 3 4 is a block diagram of the multi-scale feature extraction module in the geographic information remote sensing mapping system according to an embodiment of the present application.
[0039] Figure 4 Flow chart of a geographic information remote sensing mapping method according to an embodiment of the present application.
[0040] Figure 5 Schematic diagram of the system architecture of the geographic information remote sensing mapping method according to an embodiment of the present application. DETAILED DESCRIPTION
[0041] The following will describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0042] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present application have the same meaning as those commonly understood by those skilled in the art of the present application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application.
[0043] In the embodiments of the present application, it should be noted that, unless otherwise specified and limited, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection or a connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.
[0044] It should be noted that the terms "first\second\third" involved in the embodiments of the present application are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\third" can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\third" can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.
[0045] As mentioned above, at present, when classifying remote sensing images, due to the existence of problems such as different spectra for the same object and the same spectra for different objects, there are several major problems in supervised classification: the problem of mixing water bodies and mountain shadows; the unclear classification boundary between unused land and urban public transportation land; rivers will be misjudged as urban public transportation land; confusion between forest land and cultivated land; mountain shadows will be misjudged as urban public transportation land. Therefore, an optimized geographic information remote sensing mapping system is expected.
[0046] Accordingly, considering that when classifying remote sensing images for actual remote sensing mapping of geographic information, the characteristics of the target objects in the remote sensing image are the reflection of the radiation differences of the electromagnetic waves of the objects on the remote sensing image, in order to be able to identify the type, nature, spatial position, shape, size and other attributes of the objects based on the features of the objects on the remote sensing image, it is necessary to accurately analyze and identify the different implicit feature information of each object in the remote sensing image of the geographic information, so as to obtain accurate classification results of the remote sensing image. However, due to the large amount of information in the remote sensing image, the traditional method cannot accurately capture and identify the features of each object in the remote sensing image, thereby reducing the accuracy of classification. Therefore, in the technical solution of the present application, based on the spectral characteristics of the objects, the implicit feature information of the objects in the remote sensing spectral image is extracted, and all the pixels of the image are divided into several categories according to their properties. In this process, the difficulty lies in how to fully and accurately express the implicit feature distribution information about the objects in the remote sensing spectral image, so as to accurately identify and classify the objects, so as to accurately divide the remote sensing spectral image into several object category sub-regions with different pixel properties.
[0047] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.
[0048] The development of deep learning and neural networks has provided new solutions and ideas for mining the implicit feature distribution information of objects in remote sensing spectral images.
[0049] Specifically, in the technical solution of the present application, first, a remote sensing multispectral image to be processed is obtained. Then, considering that the implicit feature distribution information about various types of objects in the remote sensing multispectral image to be processed will be presented in the image brightness, color and texture of the remote sensing multispectral image to be processed, in order to accurately express the features of various types of objects in the remote sensing multispectral image to be processed, in the technical solution of the present application, the directional gradient histogram of the remote sensing multispectral image to be processed is calculated, and then the multi-channel image obtained by aggregating the directional gradient histogram and the remote sensing multispectral image to be processed along the channel dimension is used as input data to further improve the feature expression capability of the various types of objects.
[0050] It should be understood that here, the directional gradient histogram is a method for describing the local brightness, color and texture features of an image. The algorithm divides the image into small-sized cell spaces, calculates the gradient of each pixel in the cell, and generates cells (HOG, Histogram of Oriented Gradient) according to the gradient distribution. Then, the HOG distribution of each cell is statistically analyzed in a larger block space to generate a block space HOG to describe the local texture information. Therefore, using the multi-channel image data obtained by aggregating the directional gradient histogram and the remote sensing multispectral image to be processed as input data can enable the feature extractor to more effectively and easily extract the brightness, color and texture information related features of the various types of objects in the image, and the category features of the objects are often reflected in the brightness, color and texture information level of the image.
[0051] Further, considering that in the actual process of detecting and classifying the category of the object, the implicit features of the object in the image are small-scale feature information relative to the multi-channel image, it is difficult to capture and extract the effective features of the object in the image, thereby reducing the expression ability of the features of the object in the multi-channel image. Therefore, in the technical solution of the present application, the multi-channel image is further subjected to image block processing to obtain a sequence of image blocks. Accordingly, in a specific example of the present application, the multi-channel image can be subjected to uniform image block processing to facilitate the subsequent improvement of the implicit feature expression ability of the various types of objects, thereby obtaining a sequence of image blocks, where each image block in the sequence of image blocks has the same size. It should be understood that after the image block processing, the scale of each image block in the sequence of image blocks is reduced compared to the original image, so that the distribution information of the implicit features of the small-scale objects in the multi-channel image is no longer a small-size object in the image block, so as to improve the accurate expression of the implicit features of the various types of objects, thereby improving the classification accuracy of the various types of objects.
[0052] Then, a convolutional neural network model with excellent performance in extracting hidden features of images is used to mine the features of the sequence of image blocks. In particular, considering that when identifying and classifying the various types of land objects, the implicit feature information about the various types of land objects in each image block of the sequence of image blocks not only has large-scale features, but also has small-scale features. Therefore, in order to improve the expression ability of the hidden features of the various types of land objects, so as to accurately classify the various types of land objects, in the technical solution of the present application, a convolutional neural network model with different feature receptive fields is further used to extract the hidden features of each image block respectively. That is, specifically, in the technical solution of the present application, the sequence of image blocks is respectively passed through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices. It is worth mentioning that here, the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales to extract the multi-scale implicit feature distribution information about the various types of land objects in each image block.
[0053] Next, considering that there is a correlation between the multi-scale implicit feature distribution information about various types of land objects in each image block, in the technical solution of the present application, the sequence of the image block feature matrices is further arranged into a multi-channel global feature matrix to fuse the multi-scale implicit feature information about various types of land objects in each image block, that is, the land object category feature information about the properties of different categories of pixels in the remote sensing multispectral image to be processed.
[0054] Furthermore, after obtaining the feature information of the ground object category of different categories of pixel properties in the remote sensing multispectral image to be processed, the multi-channel global feature matrix is subjected to image semantic segmentation to obtain an image semantic segmentation result. Specifically, in the technical solution of the present application, the image semantic segmentation result is to segment the remote sensing multispectral image to be processed into several ground object category sub-regions of different categories of pixel properties, so as to accurately identify and classify the ground objects.
[0055] In particular, in the technical solution of the present application, here, when the sequence of the image blocks is respectively obtained by a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of the image block feature matrices, each image block feature matrix in the sequence of the image block feature matrices is obtained by fusion of a first branch feature matrix obtained by the first convolutional neural network model and a second branch feature matrix obtained by the second convolutional neural network model. In order to improve the fusion effect of the first branch feature matrix and the second branch feature matrix, in the technical solution of the present application, convolutional dictionary comparison response learning is preferably used to perform feature fusion, which is expressed as:
[0056]
[0057] Where M1 and M2 are the first branch feature matrix and the second branch feature matrix respectively, and ‖·‖ F Represents the Frobenius norm of a matrix.
[0058] That is, based on the neighborhood operator properties of the convolution kernel representation of the convolutional neural network of the first branch feature matrix M1 and the second branch feature matrix M2, the n-level (n-hop) neighbors of the eigenvalues of the first branch feature matrix M1 and the second branch feature matrix M2 are compared and learned based on the spatial structured convolutional dictionary comparison of the point-plus-feature flow between the corresponding features to express the features of the intrinsic prior structure, and the prior knowledge represented by the low-rank fusion is used as a feature response reference for the high-dimensional feature distribution, so as to learn the interpretable response between the features, thereby improving the image block feature matrix M obtained after fusion. c The feature expression effect is improved, thereby improving the accuracy of the image semantic segmentation result of the multi-channel global feature matrix obtained by the sequence reconstruction of the image block feature matrix. In this way, the identification and classification of objects can be accurately performed to accurately segment the remote sensing spectral image into several object category sub-regions with different pixel properties.
[0059] Figure 1 FIG. 1 is an application scenario diagram of a geographic information remote sensing mapping system according to an embodiment of the present application. Figure 1 As shown, in this application scenario, first, the remote sensing multispectral image to be processed is obtained (for example, Figure 1 Then, the acquired remote sensing multispectral image to be processed is input to a server (for example, Figure 1 In S) shown in , the server is capable of processing the remote sensing multispectral image to be processed based on a geographic information remote sensing mapping algorithm to perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result.
[0060] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.
[0061] In one embodiment of the present application, Figure 2 FIG. 1 is a block diagram of a geographic information remote sensing mapping system according to an embodiment of the present application. Figure 2As shown, according to the embodiment of the present application, the geographic information remote sensing mapping system 100 includes: a multispectral image acquisition module 110, which is used to obtain a remote sensing multispectral image to be processed; a directional gradient histogram transformation module 120, which is used to calculate the directional gradient histogram of the remote sensing multispectral image to be processed; a multi-channel aggregation module 130, which is used to aggregate the remote sensing multispectral image to be processed and the directional gradient histogram along the channel dimension to obtain a multi-channel image; an image blocking module 140, which is used to perform image blocking processing on the multi-channel image to obtain a sequence of image blocks; a multi-scale feature extraction module 150, which is used to pass the sequence of image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales; a global feature association module 160, which is used to arrange the sequence of image block feature matrices into a multi-channel global feature matrix; and a semantic segmentation module 170, which is used to perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result.
[0062] Specifically, in the embodiment of the present application, the multispectral image acquisition module 110 is used to obtain the remote sensing multispectral image to be processed. As mentioned above, at present, when classifying remote sensing images, due to the existence of problems such as different spectra for the same object and the same spectra for different objects, there are several major problems in supervised classification: the problem of mixing water bodies and mountain shadows; the unclear classification boundary between unused land and urban public transportation land; rivers will be misjudged as urban public transportation land; confusion between forest land and cultivated land; mountain shadows will be misjudged as urban public transportation land. Therefore, an optimized geographic information remote sensing mapping system is desired.
[0063] Accordingly, considering that when classifying remote sensing images for actual remote sensing mapping of geographic information, the characteristics of the target objects in the remote sensing image are the reflection of the radiation differences of the electromagnetic waves of the objects on the remote sensing image, in order to be able to identify the type, nature, spatial position, shape, size and other attributes of the objects based on the features of the objects on the remote sensing image, it is necessary to accurately analyze and identify the different implicit feature information of each object in the remote sensing image of the geographic information, so as to obtain accurate classification results of the remote sensing image. However, due to the large amount of information in the remote sensing image, the traditional method cannot accurately capture and identify the features of each object in the remote sensing image, thereby reducing the accuracy of classification. Therefore, in the technical solution of the present application, based on the spectral characteristics of the objects, the implicit feature information of the objects in the remote sensing spectral image is extracted, and all the pixels of the image are divided into several categories according to their properties. In this process, the difficulty lies in how to fully and accurately express the implicit feature distribution information about the objects in the remote sensing spectral image, so as to accurately identify and classify the objects, so as to accurately divide the remote sensing spectral image into several object category sub-regions with different pixel properties.
[0064] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.
[0065] The development of deep learning and neural networks has provided new solutions and ideas for mining the implicit feature distribution information of objects in remote sensing spectral images.
[0066] Specifically, in the technical solution of the present application, first, a remote sensing multispectral image to be processed is obtained.
[0067] Specifically, in the embodiment of the present application, the directional gradient histogram transformation module 120 is used to calculate the directional gradient histogram of the remote sensing multispectral image to be processed. Then, considering that the implicit feature distribution information about various types of objects in the remote sensing multispectral image to be processed will be presented in the image brightness, color and texture ends of the remote sensing multispectral image to be processed, in order to accurately express the features of various types of objects in the remote sensing multispectral image to be processed, in the technical solution of the present application, the directional gradient histogram of the remote sensing multispectral image to be processed is calculated, and then the multi-channel image obtained by aggregating the directional gradient histogram and the remote sensing multispectral image to be processed along the channel dimension is used as input data to further improve the feature expression capability of the various types of objects.
[0068] Specifically, in the embodiment of the present application, the multi-channel aggregation module 130 is used to aggregate the remote sensing multi-spectral image to be processed and the directional gradient histogram along the channel dimension to obtain a multi-channel image. It should be understood that here, the directional gradient histogram is a method for describing the local brightness, color and texture features of an image. The algorithm image divides the image into small-sized cell spaces, calculates the gradient of each pixel in the cell, and generates cells (HOG, Histogram of Oriented Gradient) according to the gradient distribution. Then, the HOG distribution of each cell is counted in a larger block space to generate a block space HOG to describe the local texture information. Therefore, using the multi-channel image data obtained by aggregating the directional gradient histogram and the remote sensing multi-spectral image to be processed as input data can enable the feature extractor to more effectively and easily extract the brightness, color and texture information related features of the various types of objects in the image, and the category features of the objects are often reflected in the brightness, color and texture information level of the image.
[0069] Specifically, in the embodiment of the present application, the image segmentation module 140 is used to perform image segmentation processing on the multi-channel image to obtain a sequence of image blocks. Furthermore, considering that in the actual process of detecting and classifying the category of the object, the implicit features of the object in the image are small-scale feature information relative to the multi-channel image, it is difficult to capture and extract the effective features of the object in the image, thereby reducing the expression ability of the multi-channel image about the features of the object. Therefore, in the technical solution of the present application, the multi-channel image is further subjected to image segmentation processing to obtain a sequence of image blocks.
[0070] Accordingly, in a specific example of the present application, the multi-channel image can be processed by uniform image segmentation, so as to improve the ability to express the implicit features of the various types of objects in the subsequent process, thereby obtaining a sequence of image blocks, where each image block in the sequence of image blocks has the same size. It should be understood that after the image segmentation process, the scale of each image block in the sequence of image blocks is reduced compared to the original image, so that the distribution information of the implicit features of small-scale objects in the multi-channel image is no longer a small-size object in the image block, so as to improve the accurate expression of the implicit features of the various types of objects, and thus improve the classification accuracy of the various types of objects.
[0071] Specifically, in the embodiment of the present application, the multi-scale feature extraction module 150 is used to pass the sequence of the image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales. Then, a convolutional neural network model with excellent performance in implicit feature extraction of images is used to perform feature mining of the sequence of image blocks, in particular, considering that when identifying and classifying the various types of land objects, the implicit feature information about the various types of land objects in each image block of the sequence of image blocks has not only large-scale features, but also small-scale features.
[0072] Therefore, in order to improve the expression ability of the hidden features of the various types of landforms so as to accurately classify the various types of landforms, in the technical solution of the present application, convolutional neural network models with different feature receptive fields are further used to extract the hidden features of each image block respectively. That is, specifically, in the technical solution of the present application, the sequence of image blocks is respectively passed through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices. It is worth mentioning that here, the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales to extract the multi-scale implicit feature distribution information about the various types of landforms in each image block.
[0073] in, Figure 3 is a block diagram of the multi-scale feature extraction module in the geographic information remote sensing mapping system according to an embodiment of the present application, such as Figure 3 As shown, the multi-scale feature extraction module 150 includes: a first-scale image feature extraction unit 151, used to use the first convolutional neural network model of the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix; a second-scale image feature extraction unit 152, used to use the second convolutional neural network model of the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a second branch feature matrix; and a multi-scale feature fusion unit 153, used to fuse the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix.
[0074] It is worth mentioning that compared with traditional feature engineering, the multi-scale neighborhood feature extraction module is essentially a deep neural network model based on deep learning, which can fit any function through a predetermined training strategy and has higher feature extraction generalization ability.
[0075] The multi-scale neighborhood feature extraction module includes multiple parallel one-dimensional convolutional layers, wherein, during the feature extraction process of the multi-scale neighborhood feature extraction module, the multiple parallel one-dimensional convolutional layers respectively perform one-dimensional convolution encoding on the input data with one-dimensional convolution kernels with different scales to capture local implicit features of the sequence.
[0076] Among them, the first-scale image feature extraction unit 151 is used to: use each layer of the first convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer so as to output the first branch feature matrix from the last layer of the first convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the first convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
[0077] Furthermore, the second-scale image feature extraction unit 152 is used to: use each layer of the second convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer to output the second branch feature matrix from the last layer of the second convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the second convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
[0078] In particular, in the technical solution of the present application, here, when the sequence of the image blocks is respectively obtained by a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of the image block feature matrices, each image block feature matrix in the sequence of the image block feature matrices is obtained by fusion of a first branch feature matrix obtained by the first convolutional neural network model and a second branch feature matrix obtained by the second convolutional neural network model. In order to improve the fusion effect of the first branch feature matrix and the second branch feature matrix, in the technical solution of the present application, convolutional dictionary control response learning is preferably used to perform feature fusion, which is expressed as: convolutional dictionary control response learning is used to fuse the first branch feature matrix and the second branch feature matrix with the following optimization formula to obtain the image block feature matrix; wherein the optimization formula is:
[0079]
[0080] Where M1 and M2 are the first branch feature matrix and the second branch feature matrix respectively, and ‖·‖ F represents the Frobenius norm of the matrix, represents matrix addition, Represents matrix multiplication, M c represents the image block feature matrix, (·) T Represents the transpose of a matrix.
[0081] That is, based on the neighborhood operator properties of the convolution kernel representation of the convolutional neural network of the first branch feature matrix M1 and the second branch feature matrix M2, the n-level (n-hop) neighbors of the eigenvalues of the first branch feature matrix M1 and the second branch feature matrix M2 are compared and learned based on the spatial structured convolutional dictionary comparison of the point-plus-feature flow between the corresponding features to express the features of the intrinsic prior structure, and the prior knowledge represented by the low-rank fusion is used as a feature response reference for the high-dimensional feature distribution, so as to learn the interpretable response between the features, thereby improving the image block feature matrix M obtained after fusion. c The feature expression effect is improved, thereby improving the accuracy of the image semantic segmentation result of the multi-channel global feature matrix obtained by the sequence reconstruction of the image block feature matrix. In this way, the identification and classification of objects can be accurately performed to accurately segment the remote sensing spectral image into several object category sub-regions with different pixel properties.
[0082] Specifically, in the embodiment of the present application, the global feature association module 160 is used to arrange the sequence of the image block feature matrix into a multi-channel global feature matrix. Next, considering that the multi-scale implicit feature distribution information about various types of objects in each image block has a correlation relationship, therefore, in the technical solution of the present application, the sequence of the image block feature matrix is further arranged into a multi-channel global feature matrix, so as to fuse the multi-scale implicit feature information about various types of objects in each image block, that is, the object category feature information about the properties of different categories of pixels in the remote sensing multispectral image to be processed.
[0083] The global feature association module 160 is used to arrange the sequence of the image block feature matrix into a multi-channel global feature matrix according to the following arrangement formula; wherein the arrangement formula is:
[0084] M c =Concat[M1, M2…M n ]
[0085] Among them, M1, M2…M n represents the sequence of the image block feature matrix, Concat[·] represents the cascade function, M c represents the multi-channel global feature matrix.
[0086] Specifically, in an embodiment of the present application, the semantic segmentation module 170 is used to perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result. Furthermore, after obtaining the feature information of the land object category about the properties of different categories of pixels in the remote sensing multispectral image to be processed, image semantic segmentation is performed on the multi-channel global feature matrix to obtain an image semantic segmentation result. Specifically, in the technical solution of the present application, the image semantic segmentation result is to divide the remote sensing multispectral image to be processed into several land object category sub-regions of different categories of pixel properties, so as to accurately identify and classify the land objects.
[0087] In a specific example of the present application, the image semantic segmentation result is to segment the remote sensing multispectral image to be processed into several ground object category sub-regions with different categories of pixel properties.
[0088] In summary, the geographic information remote sensing mapping system 100 based on the embodiment of the present application is explained, which obtains the remote sensing multispectral image to be processed; uses artificial intelligence technology based on deep learning to mine the spectral features of the objects in the remote sensing spectral image, and extracts the implicit feature information of the objects in the remote sensing spectral image based on the spectral features of the objects, so as to classify all the pixels of the image into several categories according to their properties. In this way, the objects can be accurately identified and classified, so as to accurately segment the remote sensing spectral image into several object category sub-regions with different pixel properties.
[0089] As described above, the geographic information remote sensing mapping system 100 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for geographic information remote sensing mapping. In one example, the geographic information remote sensing mapping system 100 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the geographic information remote sensing mapping system 100 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 geographic information remote sensing mapping system 100 can also be one of the many hardware modules of the terminal device.
[0090] Alternatively, in another example, the geographic information remote sensing mapping system 100 and the terminal device may also be separate devices, and the geographic information remote sensing mapping system 100 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.
[0091] In one embodiment of the present application, Figure 4 FIG. 1 is a flow chart of a method for mapping geographic information remote sensing according to an embodiment of the present application. Figure 4As shown, according to the geographic information remote sensing mapping method of the embodiment of the present application, it includes: 210, obtaining a remote sensing multispectral image to be processed; 220, calculating a directional gradient histogram of the remote sensing multispectral image to be processed; 230, aggregating the remote sensing multispectral image to be processed and the directional gradient histogram along the channel dimension to obtain a multi-channel image; 240, performing image block processing on the multi-channel image to obtain a sequence of image blocks; 250, respectively passing the sequence of image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales; 260, arranging the sequence of image block feature matrices into a multi-channel global feature matrix; and, 270, performing image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result.
[0092] Figure 5 Schematic diagram of the system architecture of the geographic information remote sensing mapping method according to an embodiment of the present application. Figure 5 As shown, in the system architecture of the geographic information remote sensing mapping method, first, a remote sensing multispectral image to be processed is obtained; then, a directional gradient histogram of the remote sensing multispectral image to be processed is calculated; then, the remote sensing multispectral image to be processed and the directional gradient histogram are aggregated along the channel dimension to obtain a multi-channel image; then, the multi-channel image is subjected to image block processing to obtain a sequence of image blocks; then, the sequence of image blocks is respectively passed through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales; then, the sequence of image block feature matrices is arranged into a multi-channel global feature matrix; and, finally, image semantic segmentation is performed on the multi-channel global feature matrix to obtain an image semantic segmentation result
[0093] In a specific example, in the above-mentioned geographic information remote sensing mapping method, the sequence of image blocks is respectively passed through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales, including: using the first convolutional neural network model with the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix; using the second convolutional neural network model with the dual-branch structure to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a second branch feature matrix; and, fusing the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix.
[0094] In a specific example, in the above-mentioned geographic information remote sensing mapping method, the first convolutional neural network model with the dual-branch structure is used to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix, including: using each layer of the first convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension, and nonlinear activation processing on the input data in the forward pass of the layer to output the first branch feature matrix from the last layer of the first convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the first convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
[0095] In a specific example, in the above-mentioned geographic information remote sensing mapping method, the second convolutional neural network model with the dual-branch structure is used to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a second branch feature matrix, including: using each layer of the second convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension, and nonlinear activation processing on the input data in the forward pass of the layer to output the second branch feature matrix from the last layer of the second convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the second convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
[0096] In a specific example, in the above-mentioned geographic information remote sensing mapping method, fusing the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix includes: using convolutional dictionary control response learning to fuse the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix using the following optimization formula; wherein the optimization formula is:
[0097]
[0098] Where M1 and M2 are the first branch feature matrix and the second branch feature matrix respectively, and ‖·‖ F represents the Frobenius norm of the matrix, represents matrix addition, Represents matrix multiplication, M c represents the image block feature matrix, (·) T Represents the transpose of a matrix.
[0099] In a specific example, in the above-mentioned geographic information remote sensing mapping method, arranging the sequence of the image block feature matrix into a multi-channel global feature matrix includes: arranging the sequence of the image block feature matrix into a multi-channel global feature matrix according to the following arrangement formula; wherein the arrangement formula is:
[0100] M c =Concat[M1, M2…M n ]
[0101] Among them, M1, M2…M n represents the sequence of the image block feature matrix, Concat[·] represents the cascade function, M c represents the multi-channel global feature matrix.
[0102] In a specific example, in the above-mentioned geographic information remote sensing mapping method, the image semantic segmentation result is to segment the remote sensing multispectral image to be processed into several ground object category sub-regions with different categories of pixel properties.
[0103] Those skilled in the art will appreciate that the specific operations of each step in the above-mentioned geographic information remote sensing mapping method have been described in detail above. Figures 1 to 3 The description of the geographic information remote sensing mapping system has been described in detail, and therefore, its repeated description will be omitted.
[0104] The present application also provides a computer program product, which includes instructions. When the instructions are executed, the device performs operations corresponding to the above method.
[0105] In one embodiment of the present application, a computer-readable storage medium is further provided, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned method.
[0106] It should be understood that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware can be adopted. Moreover, the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code can be adopted.
[0107] The method, system, and computer program product of the embodiments of the present application are described by flowcharts and / or block diagrams. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 one or more processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0108] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 one or more processes and / or boxes Figure 1 A function specified in one or more boxes.
[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 one or more processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0110] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.
[0111] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.
[0112] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.
[0113] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
[0114] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.
[0115] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.
Claims
1. A geographic information remote sensing mapping system, characterized in that: include: A multispectral image acquisition module is used to obtain remote sensing multispectral images to be processed; A directional gradient histogram transformation module, used for calculating the directional gradient histogram of the remote sensing multispectral image to be processed; A multi-channel aggregation module, used for aggregating the remote sensing multispectral image to be processed and the directional gradient histogram along the channel dimension to obtain a multi-channel image; An image block module, used for performing image block processing on the multi-channel image to obtain a sequence of image blocks; A multi-scale feature extraction module, used for respectively passing the sequence of image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales; A global feature association module, used for arranging the sequence of image block feature matrices into a multi-channel global feature matrix; and The semantic segmentation module is used to perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result.
2. The geographic information remote sensing mapping system according to claim 1, characterized in that: The multi-scale feature extraction module comprises: A first scale image feature extraction unit, configured to perform deep convolution encoding on each image block in the sequence of image blocks using the first convolutional neural network model with a dual-branch structure to obtain a first branch feature matrix; A second scale image feature extraction unit, configured to perform deep convolution encoding on each image block in the sequence of image blocks using the second convolutional neural network model with a dual-branch structure to obtain a second branch feature matrix; and A multi-scale feature fusion unit is used to fuse the first branch feature matrix and the second branch feature matrix to obtain the image block feature matrix.
3. The geographic information remote sensing mapping system according to claim 2, characterized in that: The first-scale image feature extraction unit is used to: use each layer of the first convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer so as to output the first branch feature matrix from the last layer of the first convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the first convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
4. The geographic information remote sensing mapping system according to claim 3, characterized in that: The second-scale image feature extraction unit is used to: use each layer of the second convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension and nonlinear activation processing on the input data in the forward pass of the layer to output the second branch feature matrix from the last layer of the second convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the second convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
5. The geographic information remote sensing mapping system according to claim 4, characterized in that: The multi-scale feature fusion unit is used to: adopt convolutional dictionary control response learning to fuse the first branch feature matrix and the second branch feature matrix according to the following optimization formula to obtain the image block feature matrix; Wherein, the optimization formula is: Where M1 and M2 are the first branch feature matrix and the second branch feature matrix respectively, and ‖·‖ F represents the Frobenius norm of the matrix, represents matrix addition, Represents matrix multiplication, M c represents the image block feature matrix, (·) T Represents the transpose of a matrix.
6. The geographic information remote sensing mapping system according to claim 5, characterized in that: The global feature association module is used to: arrange the sequence of the image block feature matrix into a multi-channel global feature matrix according to the following arrangement formula; Wherein, the arrangement formula is: M c =Concat[M1,M2…M n ] Among them, M1, M2…M n represents the sequence of the image block feature matrix, Concat[·] represents the cascade function, M c represents the multi-channel global feature matrix.
7. The geographic information remote sensing mapping system according to claim 6, characterized in that: The image semantic segmentation result is to segment the remote sensing multispectral image to be processed into several ground object category sub-regions with different categories of pixel properties.
8. A method for remote sensing mapping of geographic information, characterized in that: include: Acquire remote sensing multispectral images to be processed; Calculating a directional gradient histogram of the remote sensing multispectral image to be processed; Aggregating the remote sensing multispectral image to be processed and the oriented gradient histogram along a channel dimension to obtain a multi-channel image; Performing image block processing on the multi-channel image to obtain a sequence of image blocks; Passing the sequence of image blocks through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales; Arranging the sequence of the image block feature matrices into a multi-channel global feature matrix; as well as Perform image semantic segmentation on the multi-channel global feature matrix to obtain an image semantic segmentation result.
9. The method for geographical information remote sensing mapping according to claim 8, characterized in that: The sequence of image blocks is passed through a dual-branch structure including a first convolutional neural network model and a second convolutional neural network model to obtain a sequence of image block feature matrices, wherein the first convolutional neural network model and the second convolutional neural network model use two-dimensional convolution kernels with different scales, including: Using the first convolutional neural network model of the dual-branch structure, deep convolution encoding is performed on each image block in the sequence of image blocks to obtain a first branch feature matrix; Using the second convolutional neural network model with the dual-branch structure, deep convolution encoding is performed on each image block in the sequence of image blocks to obtain a second branch feature matrix; and The first branch feature matrix and the second branch feature matrix are fused to obtain the image block feature matrix.
10. The method for geographical information remote sensing mapping according to claim 9, characterized in that: The first convolutional neural network model with the dual-branch structure is used to perform deep convolution encoding on each image block in the sequence of image blocks to obtain a first branch feature matrix, including: using each layer of the first convolutional neural network model with the dual-branch structure to perform convolution processing based on a two-dimensional convolution kernel, pooling processing along the channel dimension, and nonlinear activation processing on the input data in the forward pass of the layer to output the first branch feature matrix from the last layer of the first convolutional neural network model with the dual-branch structure, wherein the input of the first layer of the first convolutional neural network model with the dual-branch structure is each image block in the sequence of image blocks.
