Wetland water body remote sensing recognition system based on large model analysis

Abstract

The invention relates to the technical field of remote sensing image analysis, in particular to a wetland water body remote sensing recognition system based on large model analysis, which comprises a remote sensing acquisition module, a feature region module, a weight analysis module and a remote sensing recognition module, the feature region module is used for carrying out region division to obtain a plurality of feature regions, obtaining the confidence of the feature regions and obtaining the relative reflectivity of the near-infrared feature regions; the weight analysis module is used for acquiring a matching group and performing weight analysis to determine a fusion weight; and the remote sensing identification module is used for carrying out image fusion by utilizing the fusion weight so as to carry out wetland water body identification. According to the invention, the identification accuracy of the wetland water body is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image analysis, and particularly relates to a wetland water body remote sensing identification system based on large model analysis. BACKGROUND

[0002] As an important part of the earth's ecological system, wetlands play multiple functions such as regulating climate, purifying water quality, providing habitats for wildlife, and protecting soil and water. Through remote sensing images, information such as the distribution, area, and changes of wetland water bodies can be obtained in a relatively short period of time. In order to better process these large-scale, high-dimensional remote sensing data, large model technology has gradually become one of the key technologies for wetland water body remote sensing identification. Large models can efficiently extract spatial and spectral features from images, improving the accuracy and robustness of wetland water body identification.

[0003] When the current large model identifies water bodies in remote sensing images, it will evenly distribute "attention" to all areas in the image. However, since wetland water bodies often coexist with surrounding vegetation, soil, and other water bodies, the boundaries of water bodies in remote sensing images are not always clear and are often affected by complex backgrounds. Therefore, when traditional large models identify, complex backgrounds such as grass, buildings, and roads may be misidentified as water bodies, reducing the accuracy of water body identification. SUMMARY

[0004] The present application provides a wetland water body remote sensing identification system based on large model analysis to solve the existing problems.

[0005] The wetland water body remote sensing identification system based on large model analysis of the present application adopts the following technical solutions: In a first aspect, an embodiment of the present application provides a wetland water body remote sensing identification system based on large model analysis, which includes the following modules: A remote sensing acquisition module for acquiring remote sensing images and ground reflectivity, the remote sensing images including visible light remote sensing images and near-infrared remote sensing images; A feature region module for dividing the remote sensing image region to obtain a plurality of feature regions, the feature regions being divided into visible light feature regions and near-infrared feature regions, the confidence of the feature regions being obtained using the gray scale distribution and texture features, and the contour shape features of the feature regions; the relative reflectivity of the near-infrared feature regions being obtained using the ground reflectivity within the near-infrared feature regions; A weight analysis module for matching the visible light feature regions and the near-infrared feature regions at the same time to form a matching group, and performing weight analysis based on the confidence of the visible light feature regions and the relative reflectivity of the near-infrared feature regions in the matching group to determine the fusion weight; The remote sensing recognition module is configured to perform image fusion on the feature regions in the matching group by using a fusion weight, take the fusion result as an input of a neural network, and output a wetland water body recognition result.

[0006] Optionally, the division of the remote sensing image region into a plurality of feature regions comprises the following specific method: For the visible light remote sensing image and the near-infrared remote sensing image, the ISODATA algorithm is used to cluster the pixel points in the images, a plurality of clustering clusters are obtained in the visible light remote sensing image and the near-infrared remote sensing image, each clustering cluster includes a region of pixel points in the image, which is denoted as a feature region, the feature region in the visible light remote sensing image is denoted as a visible light feature region, and the feature region in the near-infrared remote sensing image is denoted as a near-infrared feature region.

[0007] Optionally, the confidence of the feature region is obtained by using the gray scale distribution, the texture feature, and the contour shape feature of the feature region, and the specific method comprises the following steps: For any visible light feature region in the visible light remote sensing image, a gray scale co-occurrence matrix of the visible light feature region is obtained, and the entropy values of all elements in the gray scale co-occurrence matrix are taken as texture information entropy of the visible light feature region; the confidence of the visible light feature region in the visible light remote sensing image is obtained according to the difference between the average gray scale value of the visible light feature region and the average gray scale value of all visible light feature regions, the texture information entropy of the visible light feature region, and the fractal dimension of the visible light feature region, the difference is positively correlated with the confidence, and the texture information entropy and the fractal dimension are negatively correlated with the confidence; the confidence of any near-infrared feature region in the near-infrared remote sensing image is obtained, and the confidence of the near-infrared feature region is obtained in the same way as the confidence of the visible light feature region.

[0008] Optionally, the relative reflectivity of the near-infrared feature region is obtained by using the ground reflectivity in the near-infrared feature region, and the specific method comprises the following steps: A reflectivity threshold is preset, the ground reflectivity of any near-infrared feature region in the near-infrared remote sensing image is obtained, and the reflectivity threshold and the ground reflectivity of the near-infrared feature region in the near-infrared remote sensing image are normalized to obtain the relative reflectivity of the near-infrared feature region.

[0009] Optionally, the visible light feature region and the near-infrared feature region at the same time are matched to form a matching group, and the specific method comprises the following steps: The matching degree of the visible light feature region and the near-infrared feature region is obtained by using the distance and coincidence of the visible light feature region in the visible light remote sensing image and the near-infrared feature region in the near-infrared remote sensing image at the same time, and the matching degree is used to determine a plurality of matching groups, wherein each matching group is composed of one visible light feature region and one near-infrared feature region.

[0010] Optionally, the specific method for obtaining the matching degree comprises the following steps: For the visible light remote sensing image and the near-infrared remote sensing image at the same time, the centroids of the visible light feature region in the visible light remote sensing image and the near-infrared feature region in the near-infrared remote sensing image are obtained, and the Euclidean distance between the coordinates corresponding to the centroids is obtained; the sets of coordinates of all pixel points contained in any visible light feature region and any near-infrared feature region are obtained, denoted as a first region set and a second region set, the intersection of the first region set and the second region set is obtained, and the number of elements in the intersection is taken as the coincidence degree of the visible light feature region and the near-infrared feature region corresponding to the first region set and the second region set, respectively; the Euclidean distance between the coordinates corresponding to the centroids of any visible light feature region and any near-infrared feature region and the coincidence degree are combined to calculate the matching degree of the visible light feature region and the near-infrared feature region, wherein the Euclidean distance is negatively correlated with the matching degree, and the coincidence degree is positively correlated with the matching degree.

[0011] Optionally, the weight analysis based on the confidence of the visible light feature region and the relative reflectivity of the near-infrared feature region in the matching group is used to determine the fusion weight, and the specific method comprises the following steps: the confidence of the visible light feature region is taken as the spatial attention weight of the corresponding visible light feature region; the relative reflectivity of the near-infrared feature region is taken as the spatial attention weight of the corresponding near-infrared feature region; the confidence is used to divide the water body region and the non-water body region in the remote sensing image, and the discrimination coefficient of the corresponding remote sensing image is calculated according to the spatial attention weight level difference between all water body regions and non-water body regions in the remote sensing image; For any matching group, the comprehensive spatial attention weight of the matching group is calculated according to the spatial attention weight of the visible light feature region, the discrimination coefficient of the visible light remote sensing image to which the visible light feature region belongs, the spatial attention weight of the near-infrared feature region, and the discrimination coefficient of the near-infrared remote sensing image to which the near-infrared feature region belongs; The comprehensive spatial attention weight of the same matching group at different times is obtained, and a corresponding sequence is formed in time sequence, denoted as an attention weight sequence, and the weight stability degree of each matching group is calculated by using the comprehensive spatial attention weight difference of the matching group in the attention weight sequence. obtaining the maximum comprehensive spatial attention weight in the attention weight sequence and the mean of all comprehensive spatial attention weights, obtaining the difference between the comprehensive spatial attention weight of any matching group in the attention weight sequence and the maximum comprehensive spatial attention weight and the mean, respectively, and calculating the weight stability degree of the matching group; for any attention weight sequence, obtaining the comprehensive spatial attention weight of the matching group corresponding to the maximum weight stability degree as the fusion weight.

[0012] Optionally, the specific method for obtaining the discrimination coefficient of the remote sensing image comprises the following steps: For any remote sensing image, a feature region with a confidence greater than or equal to a preset confidence threshold in the remote sensing image is regarded as a water body region, and a feature region other than the water body region is regarded as a non-water body region, the water body region and the non-water body region in the remote sensing image are obtained, the average spatial attention weight of all water body regions in the remote sensing image is obtained as the water body attention weight of the remote sensing image, the average spatial attention weight of all non-water body regions in the remote sensing image is obtained as the non-water body attention weight of the remote sensing image, and the discrimination coefficient of the remote sensing image is obtained according to the difference between the water body attention weight and the non-water body attention weight.

[0013] Optionally, the specific method for obtaining the weight stability degree of the matching group comprises the following steps: For any two matching groups at different time points, a first region set of visible light feature regions and a second region set of near-infrared feature regions contained in the two matching groups, a set further formed by the first region set and the second region set contained in any matching group is regarded as a total set of the matching group, the intersection-over-union of the total sets of any two matching groups at different time points is obtained, when the intersection-over-union is maximum, the two matching groups are the same matching group at different time points, a sequence formed by the comprehensive spatial attention weights corresponding to the same matching group at all continuous time points is obtained as an attention weight sequence; obtaining the maximum comprehensive spatial attention weight in the attention weight sequence and the mean of all comprehensive spatial attention weights, obtaining the difference between the comprehensive spatial attention weight of any matching group in the attention weight sequence and the maximum comprehensive spatial attention weight and the mean, respectively, and calculating the weight stability degree of the matching group.

[0014] Optionally, the specific method for obtaining the weight stability degree of the matching group comprises the following steps: For any attention weight sequence, the matching group corresponding to the element contained in the attention weight sequence when the weight stability degree is maximum is obtained as the target matching group, the comprehensive spatial attention weight of the target matching group is used as the fusion weight in the alpha fusion algorithm, so as to fuse the visible light feature region and the near-infrared feature region in the target matching group, and obtain a fusion region; all fusion regions are obtained and used as the input of the trained convolutional neural network, the convolutional neural network is used for wetland water body identification, and the identification result is output by the convolutional neural network.

[0015] The beneficial effects of the technical scheme of the present application are: the visible light image and the near-infrared image respectively represent different information dimensions. The visible light image reflects the visible region that can be perceived by the human eye, while the near-infrared image can reveal the reflection characteristics of different substances on the ground, especially for the identification of wetland water bodies. By fusing the two, not only can the shortcomings of a single image source be made up, but also the accuracy and robustness of wetland water body identification can be effectively improved; in addition, in the image recognition process, dividing the feature regions can effectively reduce the amount of calculation, and the data of different feature regions are processed respectively, avoiding redundant calculation on the entire image. In this way, key features can be more efficiently extracted and analyzed under limited computing resources, improving the real-time performance and computing efficiency of the system; by combining different image features (such as gray distribution, texture, and contour shape), more discriminative information can be introduced in the analysis of the feature regions, the accuracy is improved, and the confidence evaluation allows the system to adjust the judgment strategy according to the complexity of the image features, making the system more flexible and accurate; further, by matching the visible light feature region with the near-infrared feature region and determining the fusion weight through weight analysis, the scheme can perform weighted processing on the image according to the reliability and reflectivity of different regions, and then generate a more accurate fusion result, thereby effectively enhancing the accuracy of wetland water body identification in the image fusion result. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical schemes in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0017] Figure 1 The structural block diagram of the wetland water body remote sensing identification system based on large model analysis of the present application. DETAILED DESCRIPTION

[0018] To further illustrate the technical means and effectiveness of the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of the wetland water body remote sensing identification system based on large-scale model analysis proposed by the present invention. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0020] The specific scheme of the wetland water body remote sensing identification system based on large model analysis provided by the present invention is described in detail below with reference to the accompanying drawings.

[0021] See also Figure 1 , which shows a structural block diagram of a wetland water body remote sensing identification system based on large model analysis provided by one embodiment of the present invention. The system includes the following modules: The remote sensing acquisition module 101 is used to obtain remote sensing images and ground reflectivity.

[0022] It should be noted that convolutional neural networks (CNNs) are one of the most common deep learning models for processing remote sensing images. They can automatically extract spatial features from images and are used to identify the boundaries, morphology, and other characteristics of wetland water bodies. CNNs typically use labeled remote sensing imagery for supervised learning, enabling accurate water extraction. Because wetland water bodies often coexist with surrounding vegetation, soil, and other water bodies, their boundaries in remote sensing images are affected by complex backgrounds. The uniform attention mechanism used in current large-scale recognition models can lead to recognition errors for water bodies in such complex backgrounds. Therefore, this embodiment of the present invention incorporates multispectral remote sensing imagery for precise identification. Furthermore, since most water bodies (especially stagnant water or wetland water bodies) have near-zero or extremely low reflectance in the near-infrared band, while vegetation or bare land have very high reflectance in the near-infrared band, these can be more clearly distinguished.

[0023] In order to implement the wetland water body remote sensing identification system based on large model analysis proposed in this embodiment, it is first necessary to collect remote sensing images of the wetland. The specific process is as follows: Firstly, visible light remote sensing images and near-infrared remote sensing images of any wetland area are collected using a preset fixed sampling interval. Visible light remote sensing images and near-infrared remote sensing images are collectively referred to as remote sensing images, and ground reflectance data are obtained.

[0024] It should be noted that, in the embodiment of the present invention, the corresponding image resolution and position of each sampling are preset and guaranteed to be the same based on experience.

[0025] Then, the visible light remote sensing image and the near-infrared remote sensing image are filtered.

[0026] At this point, the visible light remote sensing image and the near-infrared remote sensing image are obtained through the above method.

[0027] The feature region module 102 is configured to divide the remote sensing image region to obtain a plurality of feature regions, the feature regions are divided into visible light feature regions and near-infrared feature regions, the confidence of the feature regions is obtained by using the gray distribution and the texture feature and the contour shape feature of the feature regions, and the relative reflectivity of the near-infrared feature regions is obtained by using the ground reflectivity in the near-infrared feature regions.

[0028] It should be noted that, due to the uniform attention mechanism of the current large model recognition, the water body in the above complex background may cause recognition errors, and in this embodiment, the spatial attention weight of different regions of the water body in the multispectral image of this area is determined by analyzing the difference characteristics of the water body, the regions of the multispectral image are matched, and the attention weight of the contribution degree of different matching regions to the water body is fused in time and space. The unbalanced attention mechanism described in this embodiment assigns different weights to different regions of the image, so that the network large model can concentrate on processing important regions while ignoring unimportant regions, thereby improving the accuracy of water body recognition. Therefore, first, the spatial attention weight of the multispectral image of this area is determined. In addition, for the multispectral image of the wetland water body, since there are differences in gray scale and reflectivity between the water body region and the vegetation and bare land region, it is hoped that the water body region will have a greater attention weight than other regions, and then the spatial attention weight of different regions of the image can be determined by using the difference.

[0029] Specifically, first, the visible light remote sensing image and the near-infrared remote sensing image are segmented by a clustering algorithm respectively, to obtain a plurality of visible light feature regions in the visible light remote sensing image and a plurality of near-infrared feature regions in the near-infrared remote sensing image.

[0030] As a preferred embodiment, the visible light remote sensing image and the near-infrared remote sensing image are segmented by a clustering algorithm respectively, and the specific method includes: for the visible light remote sensing image and the near-infrared remote sensing image, the ISODATA algorithm is used to cluster the pixel points in the image, to obtain a plurality of clustering clusters in the visible light remote sensing image and the near-infrared remote sensing image, each clustering cluster contains a region of pixel points in the image, which is recorded as a feature region, the feature region in the visible light remote sensing image is recorded as a visible light feature region, and the feature region in the near-infrared remote sensing image is recorded as a near-infrared feature region.

[0031] It should be noted that the water body of the general wetland is distributed between the land vegetation, and the color of the corresponding feature region of the water body in the visible light remote sensing image is darker and the color texture is more single than the bare land region. In addition, the boundary of the water body region is usually smooth and regular due to the continuity, while the boundary of the vegetation region is usually irregular.

[0032] Then, the confidence of the visible light feature region in the visible light remote sensing image is obtained by using the gray value distribution, the texture feature and the contour shape feature in the visible light remote sensing image.

[0033] As a preferred embodiment, the method for obtaining the confidence is as follows: for any visible light feature region in the visible light remote sensing image, a gray level co-occurrence matrix of the visible light feature region is obtained, and the entropy value of all elements in the gray level co-occurrence matrix is taken as the texture information entropy of the visible light feature region. The confidence of the visible light feature region in the visible light remote sensing image is obtained according to the difference between the average gray value of the visible light feature region and the average gray value of all visible light feature regions, the texture information entropy of the visible light feature region and the fractal dimension of the visible light feature region. The difference is positively correlated with the confidence, and the texture information entropy and the fractal dimension are negatively correlated with the confidence. The confidence of any near-infrared feature region in the near-infrared remote sensing image is obtained by using the same method as that for obtaining the confidence of the visible light feature region.

[0034] As an optional embodiment, the specific calculation method of the confidence of the visible light feature region is as follows: ; In the formula, A a is the confidence of the a-th feature region in the visible light remote sensing image; is the average gray value of the a-th visible light feature region, is the maximum average gray value in all visible light feature regions, H a is the texture information entropy of the a-th visible light feature region; D a is the fractal dimension of the a-th visible light feature region; and norm() represents a linear normalization function.

[0035] It should be noted that, therefore reflects the gray distribution feature of the feature region, and the greater the value, the darker and more uniform the gray of the corresponding region, and the more likely it is a water body region; the fractal dimension D aThe fractal dimension value is greater, the corresponding morphological complexity is higher, and the contour shape of the water body of the wetland usually does not have high complexity, so that the fractal dimension value is smaller, the complexity of the edge contour of the corresponding characteristic region is lower, that is, the edge is smoother, and the confidence of the corresponding characteristic region as the water body region is greater. In addition, the possibility of each region in the wetland as the water body region can be reflected in the near-infrared remote sensing image of the wetland under the condition of near-infrared remote sensing scanning, so that the confidence of the near-infrared characteristic region in the near-infrared remote sensing image is also obtained in the embodiment of the application, to be used for subsequent division of the water body region and the non-water body region in the near-infrared remote sensing image.

[0036] In addition, in order to distinguish the water body region and the non-water body region in the remote sensing image, the confidence is used as a kind of attention weight in the embodiment of the application, so that the water body part can be given enough attention degree in subsequent wetland water body identification, and the attention to other non-water body regions such as vegetation regions is relatively smaller, and for the visible light remote sensing image, the gray texture and the fractal feature can effectively describe the related features of the wetland water body in the remote sensing image obtained by visible light.

[0037] In addition, the confidence of the visible light characteristic region is used as the spatial attention weight of the corresponding visible light characteristic region.

[0038] It should be noted that for the near-infrared remote sensing image, the reflectivity difference between the water body and the vegetation is large (the reflectivity of the vegetation is usually between 40% and 60%, and the near-infrared reflectivity of the water body is usually less than 10%), and the reflectivity degree of different characteristic regions has the same trend as the spatial attention weight, so that the relative reflectivity of any characteristic region in the near-infrared remote sensing image is obtained.

[0039] Finally, a preset reflectivity threshold is obtained, the ground reflectivity of any near-infrared characteristic region in the near-infrared remote sensing image is obtained, the ratio of the reflectivity threshold to the ground reflectivity of the near-infrared characteristic region in the near-infrared remote sensing image is normalized, and the relative reflectivity of the near-infrared characteristic region is obtained.

[0040] As an optional embodiment, the specific calculation method of the relative reflectivity of the characteristic region is: ; In the formula, B b is the relative reflectivity of the bth near-infrared characteristic region in the near-infrared remote sensing image; R0 is a preset reflectivity threshold, R b is the ground reflectivity of the bth near-infrared characteristic region in the near-infrared remote sensing image; and norm() represents a linear normalization function.

[0041] It should be noted that in the embodiments of the present application, the reflectivity threshold is empirically preset to 10%, which can be adjusted according to actual conditions, and the embodiments of the present application are not limited.

[0042] It should be noted that since the water body absorbs most of the near-infrared light, the corresponding ground reflectivity will be smaller, and the value of the relative reflectivity of the corresponding near-infrared characteristic region will be larger. The larger the value is, the higher the relative reflectivity of the corresponding near-infrared characteristic region is, and thus the more likely the near-infrared characteristic region is a water body. In addition, since the relative reflectivity of different near-infrared characteristic regions in the region represents the likelihood of being a water body, the relative reflectivity can be used as the attention weight of the corresponding near-infrared characteristic region, which can better reflect the characteristics of the near-infrared characteristic region in the near-infrared, and thus better identify wetland water bodies through the near-infrared remote sensing image.

[0043] In addition, the relative reflectivity of the near-infrared characteristic region is used as the spatial attention weight of the corresponding near-infrared characteristic region.

[0044] At this point, the confidence of the visible light characteristic region in the visible light remote sensing image and the relative reflectivity of the near-infrared characteristic region in the near-infrared remote sensing image are obtained through the above method.

[0045] The weight analysis module 103 is configured to match the visible light characteristic region and the near-infrared characteristic region at the same time to form a matching group, and perform weight analysis based on the confidence of the visible light characteristic region and the relative reflectivity of the near-infrared characteristic region in the matching group to determine the fusion weight.

[0046] It should be noted that the spatial attention weights of different characteristic regions are determined for the multispectral image of the water body region through the above method. Since the information capture degree of the characteristic regions (water body, vegetation) in the region is different under different spectra, the fusion combined with the attention weight of the multispectral image can make the difference between vegetation and water body more obvious, and improve the accuracy of subsequent target recognition based on a large model. Therefore, this step performs image fusion based on the spatial attention weights of the visible light remote sensing image and the near-infrared remote sensing image of the region.

[0047] It should be noted that the attention weights of the two images in the above step are obtained independently and simultaneously, and spatial fusion requires corresponding matching (horizontal matching) of different characteristic regions in the two images at the same sampling time point. Therefore, for any two characteristic regions in the two images, if they represent the same region, the outlines and positions of the corresponding characteristic regions are more similar.

[0048] Specifically, first, the matching degree of the visible light feature region and the near-infrared feature region is obtained by using the distance and the coincidence in spatial position of the visible light feature region in the visible light remote sensing image and the near-infrared feature region in the near-infrared remote sensing image at the same time, and the matching degree is used to determine a plurality of matching groups, wherein each matching group is composed of one visible light feature region and one near-infrared feature region.

[0049] As a preferred embodiment, the specific method for obtaining the matching degree is as follows: for the visible light remote sensing image and the near-infrared remote sensing image at the same time, the centroids of the visible light feature region in the visible light remote sensing image and the near-infrared feature region in the near-infrared remote sensing image are obtained respectively, and the Euclidean distance between the coordinates corresponding to the centroids is obtained; the set of coordinates of all pixel points contained in any visible light feature region and any near-infrared feature region is obtained respectively, denoted as a first region set and a second region set, the intersection of the first region set and the second region set is obtained, and the number of elements in the intersection is taken as the coincidence degree of the visible light feature region and the near-infrared feature region corresponding to the first region set and the second region set respectively; the Euclidean distance between the coordinates corresponding to the centroids of any visible light feature region and any near-infrared feature region and the coincidence degree are combined to calculate the matching degree of the visible light feature region and the near-infrared feature region, wherein the Euclidean distance is negatively correlated with the matching degree, and the coincidence degree is positively correlated with the matching degree.

[0050] When the matching degree is greater than or equal to a preset matching degree threshold, the visible light feature region and the near-infrared feature region corresponding to the matching degree form a matching group.

[0051] It should be noted that, for the preset matching degree threshold, in the embodiment of the present application, the preset value is 0.8 according to experience, which can be adjusted according to actual conditions, and in the embodiment of the present application, the preset value is not limited.

[0052] As an optional embodiment, the specific calculation method of the matching degree is as follows: ; In the formula, Z a,b represents the matching degree of the a th visible light feature region in the visible light remote sensing image and the b th near-infrared feature region in the near-infrared remote sensing image; R a represents the first region set of the a th visible light feature region in the visible light remote sensing image; R b represents the second region set of the b th near-infrared feature region in the near-infrared remote sensing image, C a represents the centroid coordinate of the a th visible light feature region in the visible light remote sensing image; C b represents the centroid coordinate of the b th near-infrared feature region in the near-infrared remote sensing image; L( ) represents the Euclidean distance function; norm[ ] represents the linear normalization function.

[0053] It should be noted that in the matching degree obtaining method, The profile coincidence degree of two feature regions (i.e. the visible light feature region and the near-infrared feature region) in different remote sensing images is embodied, the two feature regions are superimposed in profile by the intersection method to determine the coincident part, the more the number of elements in the intersection, the greater the corresponding coincidence degree, and then the two feature regions in different remote sensing images are more likely to represent the same position region; in addition, L(C a ,C b ) represents the position proximity degree of the two feature regions, the smaller the value, the closer the position of the two feature regions in the remote sensing image, and the greater the corresponding matching degree.

[0054] It should be noted that after the visible light feature region in the visible light remote sensing image and the near-infrared feature region in the near-infrared remote sensing image at the same time are matched by the matching degree, for any matching group at any time point, the corresponding spatial attention weight of the two feature regions (i.e. the visible light feature region and the near-infrared feature region) contained exists difference, and in the weight fusion, the contribution degree of the corresponding remote sensing image to the water body identification accuracy needs to be weighted, the greater the contribution degree of the remote sensing image to the wetland water body identification, the greater the proportion of the corresponding spatial attention weight in the fusion, and vice versa.

[0055] It should be noted that since the contribution degree of a certain remote sensing image represents the ability of distinguishing water body and non-water body accuracy, the stronger the distinguishing ability, the greater the corresponding contribution degree, and in the spatial weighted fusion, the attention weight corresponding to this image is more preferred, therefore it is necessary to evaluate the distinguishing ability of different images. For two images, the visible light remote sensing image reflects the difference between water body and non-water body through the difference in gray scale, and the near-infrared remote sensing image reflects it by mapping the reflectivity difference into a gray scale image, that is, the gray scale features corresponding to the two images can represent their distinguishing ability, the greater the difference in attention weight between water body and non-water body regions in a certain image, the more obvious the image is to distinguish water body and non-water body, and the stronger the distinguishing ability.

[0056] The water body region and the non-water body region are divided in the remote sensing image by using the size of the confidence degree, and according to the spatial attention weight level difference of all water body regions and non-water body regions in the remote sensing image, the distinguishing coefficient of the corresponding remote sensing image is calculated.

[0057] As a preferred embodiment, the method for obtaining the discrimination coefficient of the remote sensing image is: for any remote sensing image, the feature area in the remote sensing image with a confidence greater than or equal to a preset confidence threshold is taken as the water body area, and the feature area outside the water body area is taken as the non-water body area, the water body area and the non-water body area in the remote sensing image are obtained, the average spatial attention weight of all water body areas in the remote sensing image is obtained as the water body attention weight of the remote sensing image, the average spatial attention weight of all non-water body areas in the remote sensing image is taken as the non-water body attention weight of the remote sensing image, and the discrimination coefficient of the remote sensing image is obtained according to the difference between the water body attention weight and the non-water body attention weight.

[0058] In a specific embodiment of the present invention, the discrimination coefficient of the remote sensing image is used to obtain the discrimination coefficients of the visible light remote sensing image and the near infrared remote sensing image respectively.

[0059] As an optional embodiment, for any remote sensing image, the specific calculation method of the discrimination coefficient of the remote sensing image is: ; Where Y is the discrimination coefficient of the remote sensing image; is the water body attention weight of the remote sensing image, is the non-water body attention weight of the remote sensing image; u represents the preset first parameter; || represents the absolute value function.

[0060] It should be noted that The larger the value of , the greater the difference between the two images, the more obvious the water area, and the stronger the discrimination ability. The discrimination coefficient of the two images at this time point is calculated using the above method. The image with stronger discrimination ability has a greater contribution to spatial fusion, and the corresponding attention weight of this image has a more significant effect in the fusion weight, while the effect is smaller.

[0061] It should be noted that in the embodiment of the present invention, the confidence threshold is preset to 0.8 based on experience. In order to avoid the denominator being 0, the first parameter is preset to 0.01. The confidence threshold and the first parameter can be adjusted according to actual conditions, and the embodiment of the present invention does not make specific requirements.

[0062] For any matching group, the comprehensive spatial attention weight of the matching group is calculated based on the spatial attention weight of the visible light feature area in the matching group and the discrimination coefficient of the visible light remote sensing image to which the visible light feature area belongs, the spatial attention weight of the near-infrared feature area and the discrimination coefficient of the near-infrared remote sensing image to which the near-infrared feature area belongs.

[0063] As an optional embodiment, for any matching group, the specific calculation method of the comprehensive spatial attention weight of the matching group is: ; Where, is the comprehensive spatial attention weight of the matching group; γk represents the spatial attention weight of the visible light feature region in the matching group; Yk represents the discrimination coefficient of the visible light remote sensing image to which the visible light feature region in the matching group belongs; γh represents the spatial attention weight of the near-infrared feature region in the matching group; Yh represents the discrimination coefficient of the near-infrared remote sensing image to which the near-infrared feature region in the matching group belongs.

[0064] It should be noted that and They respectively represent the relative discrimination coefficients of the visible light remote sensing image and the near-infrared remote sensing image compared to the two remote sensing images, reflecting the contribution of the two remote sensing images in accurately distinguishing between water body areas and non-water body areas. The attention weights of the corresponding images are weighted with them to obtain the comprehensive attention weight, and the weights are marked on the corresponding areas of the visible light remote sensing image.

[0065] It should be noted that due to the differences in feature significance of different feature areas under different sampling environments (for example, the feature distinction between water bodies and vegetation is more obvious when the sunlight is strong than in rainy weather), in order to prevent the environment from affecting the spatial attention weight of this area at a single sampling moment, further analysis is needed in combination with matching groups at different times.

[0066] It should be noted that, since the changes in wetland water bodies and vegetation land are generally relatively stable and will not change drastically within the sampling period, it is hoped that the spatial attention weight of the feature area after the fusion of visible light remote sensing images and near-infrared remote sensing images can reflect the stable characteristics of this feature area. Therefore, the embodiment of the present invention further uses remote sensing images collected at multiple times for analysis.

[0067] Obtain the comprehensive spatial attention weights of the same matching group at different times, and form a corresponding sequence in chronological order, recorded as the attention weight sequence. Use the differences in the comprehensive spatial attention weights of the matching groups in the attention weight sequence to calculate the weight stability of each matching group; for any attention weight sequence, obtain the comprehensive spatial attention weight of the corresponding matching group when the weight stability is the largest, as the fusion weight.

[0068] As a preferred embodiment, the method for obtaining the weight stability is: Firstly, for the first region set and the second region set of the visible light feature region and the near-infrared feature region contained in all matching groups at any two time points, the set further formed by the first region set and the second region set contained in any matching group is taken as the total set of the matching groups, the intersection union ratio of the total set of any two matching groups at different time points is obtained, when the intersection union ratio is maximum, the two matching groups are the same matching group at different time points, the sequence formed by the comprehensive spatial attention weight corresponding to the same matching group at all continuous time points is taken as the attention weight sequence.

[0069] Then, the maximum comprehensive spatial attention weight in the attention weight sequence and the mean of all comprehensive spatial attention weights are obtained, the difference between the comprehensive spatial attention weight of any matching group in the attention weight sequence and the maximum comprehensive spatial attention weight and the mean is obtained, and the weight stability degree of the matching group is calculated.

[0070] As an optional embodiment, for any attention weight sequence, the specific calculation method of the weight stability degree of the matching group corresponding to the element in the attention weight sequence is as follows: ; In the formula, τ t represents the weight stability degree of the matching group corresponding to the tth element in the attention weight sequence; represents the comprehensive spatial attention weight of the matching group corresponding to the tth element in the attention weight sequence; represents the mean of all comprehensive spatial attention weights in the attention weight sequence;E γ represents the maximum comprehensive spatial attention weight in the attention weight sequence;||represents the absolute value function.

[0071] It should be noted that, represents the difference between the comprehensive spatial attention weight of the tth element in the attention weight sequence and the mean of all comprehensive spatial attention weights, and the smaller the value is, the closer the weight is to the mean weight, and the higher the stability is. represents the difference between the comprehensive spatial attention weight of the tth element in the attention weight sequence and the maximum comprehensive spatial attention weight, and the larger the value is, the farther the distance from the extreme value is, and the higher the stability is. Since the maximum comprehensive spatial attention weight in the attention weight sequence represents the most unstable comprehensive spatial attention weight, in order to effectively fuse the visible light remote sensing image and the near-infrared remote sensing image and accurately identify the wetland water, the embodiment of the application selects the comprehensive spatial attention weight obtained finally to be far away from the maximum comprehensive spatial attention weight in the attention weight sequence, and the mean of all comprehensive spatial attention weights in the attention weight sequence can reflect the overall level of the attention weight sequence, so the embodiment of the application selects the comprehensive spatial attention weight obtained finally to be as close to the mean of all comprehensive spatial attention weights as possible.

[0072] At this point, the weight stability degree of the matching group is obtained by the above method.

[0073] The remote sensing recognition module 104 is configured to perform image fusion on the feature regions in the matching group by using the weight stability degree, obtain a fused region, and take the fused region as an input of a neural network to output a wetland water body recognition result.

[0074] Specifically, first, for any attention weight sequence, a matching group corresponding to the maximum weight stability degree in the elements contained in the attention weight sequence is obtained as a target matching group, and the comprehensive spatial attention weight of the target matching group is used as a fusion weight in an alpha fusion algorithm, so as to fuse the visible light feature region and the near-infrared feature region in the target matching group to obtain a fused region.

[0075] Then, all the fused regions are obtained and taken as an input of a trained convolutional neural network (CNN), and the convolutional neural network is used to recognize the wetland water body, and the convolutional neural network outputs a recognition result.

[0076] As an optional embodiment, the training process of the convolutional neural network is as follows: 1) a plurality of remote sensing images obtained in different regions and different collection environments are obtained, a plurality of fused regions are obtained through the feature region module and the weight analysis module, and each fused region is assigned a corresponding artificial label, the artificial label including a wetland water body and a non-wetland water body; 2) any one of the fused regions with the artificial label is taken as a sample, a data set formed by all the samples is obtained, and the data set is divided into a training set, a validation set and a test set according to a 6:2:2 relationship; 3) the data set is taken as an input of the convolutional neural network, the artificial label of the sample in the data set is taken as an output of the convolutional neural network, the convolutional neural network is trained, and a cross-entropy loss function is selected as a loss function of the convolutional neural network.

[0077] At this point, the embodiment is completed.

[0078] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A wetland water body remote sensing identification system based on large model analysis is characterized by: The system includes the following modules: A remote sensing acquisition module is used to obtain remote sensing images and ground reflectivity, wherein the remote sensing images include visible light remote sensing images and near infrared remote sensing images; The feature region module is used to divide the remote sensing image area into several feature regions. The feature regions are divided into visible light feature regions and near-infrared feature regions. The confidence of the feature regions is obtained by using the grayscale distribution, texture characteristics, and contour morphological characteristics of the feature regions. The relative reflectivity of the near-infrared feature regions is obtained by using the ground reflectivity within the near-infrared feature regions. A weight analysis module is used to match the visible light feature area and the near-infrared feature area at the same time to form a matching group, and perform weight analysis based on the confidence of the visible light feature area and the relative reflectivity of the near-infrared feature area in the matching group to determine the fusion weight; The remote sensing recognition module is used to perform image fusion on the feature areas in the matching group using the fusion weights, use the fusion results as the input of the neural network, and output the wetland water body recognition results.

2. The wetland water body remote sensing identification system based on large model analysis according to claim 1 is characterized in that: The specific method of dividing the remote sensing image region into several characteristic regions includes: For visible light remote sensing images and near-infrared remote sensing images, the ISODATA algorithm is used to cluster the pixels in the images, respectively. Several clusters are obtained in the visible light remote sensing images and near-infrared remote sensing images. The pixels contained in each cluster form an area in the image, which is recorded as a feature area. The feature area in the visible light remote sensing image is recorded as a visible light feature area, and the feature area in the near-infrared remote sensing image is recorded as a near-infrared feature area.

3. The wetland water body remote sensing identification system based on large model analysis according to claim 1 is characterized in that: The method of obtaining the confidence of the feature region by using the grayscale distribution and texture features and the contour morphology features of the feature region includes the following specific methods: For any visible light feature region in a visible light remote sensing image, the gray level co-occurrence matrix of the visible light feature region is obtained, and the entropy value of all elements in the gray level co-occurrence matrix is ​​used as the texture information entropy of the visible light feature region. According to the difference between the average gray value of the visible light feature region and the average gray value of all visible light feature regions, the texture information entropy of the visible light feature region and the fractal dimension of the visible light feature region, the confidence of the visible light feature region in the visible light remote sensing image is obtained, and the difference is positively correlated with the confidence, and the texture information entropy and the fractal dimension are both negatively correlated with the confidence; the confidence of any near-infrared feature region in the near-infrared remote sensing image is obtained, and the confidence of the near-infrared feature region is obtained in the same manner as the confidence of the visible light feature region.

4. The wetland water body remote sensing identification system based on large model analysis according to claim 1 is characterized in that: The method of using the ground reflectivity in the near-infrared characteristic area to obtain the relative reflectivity of the near-infrared characteristic area includes the following specific methods: A reflectivity threshold is preset, and the ground reflectivity of any near-infrared feature area in the near-infrared remote sensing image is obtained. The ratio of the reflectivity threshold to the ground reflectivity of the near-infrared feature area in the near-infrared remote sensing image is normalized to obtain the relative reflectivity of the near-infrared feature area.

5. The wetland water body remote sensing identification system based on large model analysis according to claim 1 is characterized in that: The specific method of matching the visible light characteristic region and the near infrared characteristic region at the same time to form a matching group includes: The distance and overlap in spatial position between the visible light feature area in the visible light remote sensing image and the near-infrared feature area in the near-infrared remote sensing image at the same time are used to obtain the matching degree of the visible light feature area and the near-infrared feature area, and the matching degree is used to determine several matching groups, wherein the matching group consists of a visible light feature area and a near-infrared feature area.

6. The wetland water body remote sensing identification system based on large model analysis according to claim 5 is characterized in that: The specific method for obtaining the matching degree is: For visible light remote sensing images and near-infrared remote sensing images at the same time, the centroids of the visible light feature areas in the visible light remote sensing images and the near-infrared feature areas in the near-infrared remote sensing images are obtained respectively, and the Euclidean distance between the corresponding coordinates of the centroids is obtained; the sets formed by the coordinates of all pixel points contained in any visible light feature area and any near-infrared feature area are obtained respectively, recorded as the first area set and the second area set, the intersection of the first area set and the second area set is obtained, the number of elements in the intersection is used as the overlap of the visible light feature areas and the near-infrared feature areas corresponding to the first area set and the second area set respectively, and the Euclidean distance and overlap between the corresponding coordinates of the centroids of any visible light feature area and any near-infrared feature area are combined to calculate the matching degree of the visible light feature area and the near-infrared feature area, the Euclidean distance is negatively correlated with the matching degree, and the overlap is positively correlated with the matching degree.

7. The wetland water body remote sensing identification system based on large model analysis according to claim 6 is characterized in that: The weight analysis is performed based on the confidence of the visible light feature area and the relative reflectivity of the near-infrared feature area in the matching group to determine the fusion weight, including the specific method of: The confidence of the visible light feature region is used as the spatial attention weight of the corresponding visible light feature region; The relative reflectivity of the near-infrared feature area is used as the spatial attention weight of the corresponding near-infrared feature area; The confidence level is used to divide the water area and non-water area in the remote sensing image. The discrimination coefficient of the corresponding remote sensing image is calculated based on the difference in spatial attention weight levels of all water areas and non-water areas in the remote sensing image. For any matching group, the comprehensive spatial attention weight of the matching group is calculated based on the spatial attention weight of the visible light feature area in the matching group and the discrimination coefficient of the visible light remote sensing image to which the visible light feature area belongs, the spatial attention weight of the near-infrared feature area and the discrimination coefficient of the near-infrared remote sensing image to which the near-infrared feature area belongs; Obtain the comprehensive spatial attention weights of the same matching group at different times and form a corresponding sequence in chronological order, recorded as the attention weight sequence. Use the differences in the comprehensive spatial attention weights of the matching groups in the attention weight sequence to calculate the weight stability of each matching group; Obtain the maximum comprehensive spatial attention weight and the mean of all comprehensive spatial attention weights in the attention weight sequence, obtain the difference between the comprehensive spatial attention weight of any matching group in the attention weight sequence and the maximum comprehensive spatial attention weight and the mean, and calculate the weight stability of the matching group; for any attention weight sequence, obtain the comprehensive spatial attention weight of the matching group corresponding to the maximum weight stability as the fusion weight.

8. The wetland water body remote sensing identification system based on large model analysis according to claim 7 is characterized in that: The specific method for obtaining the discrimination coefficient of the remote sensing image is: For any remote sensing image, the feature area in the remote sensing image with a confidence greater than or equal to a preset confidence threshold is taken as the water body area, and the feature area outside the water body area is taken as the non-water body area, the water body area and the non-water body area in the remote sensing image are obtained, the average spatial attention weight of all water body areas in the remote sensing image is obtained as the water body attention weight of the remote sensing image, the average spatial attention weight of all non-water body areas in the remote sensing image is taken as the non-water body attention weight of the remote sensing image, and the discrimination coefficient of the remote sensing image is obtained according to the difference between the water body attention weight and the non-water body attention weight.

9. The wetland water body remote sensing identification system based on large model analysis according to claim 7 is characterized in that: The specific method for obtaining the weight stability of the matching group is: For the first region set of visible light characteristic regions and the second region set of near-infrared characteristic regions contained in all matching groups at any two moments, a set formed by the first region set and the second region set contained in any matching group is further used as the total set of matching groups, and the intersection-over-union ratio of the total sets of any two matching groups at different moments is obtained. When the intersection-over-union ratio is maximum, the two matching groups are the same matching group at different moments. A sequence formed by the comprehensive spatial attention weights corresponding to the same matching group at all consecutive moments is obtained as the attention weight sequence; Obtain the maximum comprehensive spatial attention weight in the attention weight sequence and the mean of all comprehensive spatial attention weights, obtain the difference between the comprehensive spatial attention weight of any matching group in the attention weight sequence and the maximum comprehensive spatial attention weight and the mean, and calculate the weight stability of the matching group.

10. The wetland water body remote sensing identification system based on large model analysis according to claim 7 is characterized in that: The method of using the fusion weight to perform image fusion on the feature areas in the matching group, using the fusion result as the input of the neural network, and outputting the wetland water body recognition result includes the following specific methods: For any attention weight sequence, the matching group corresponding to the elements contained in the attention weight sequence when the weight stability is the largest is obtained as the target matching group, and the comprehensive spatial attention weight of the target matching group is used as the fusion weight in the alpha fusion algorithm, so as to fuse the visible light feature area and the near-infrared feature area in the target matching group to obtain the fusion area; all the fusion areas are obtained and used as the input of the trained convolutional neural network, and the convolutional neural network is used to identify wetland water bodies, and the convolutional neural network outputs the recognition results.

Images