Multi-scale image segmentation and reconstruction method and system for remote sensing images

Abstract

The present invention discloses a multi-scale image segmentation and reconstruction method and system for remote sensing images, which relates to the field of image segmentation and reconstruction. The method comprises the following steps: obtaining a high-definition remote sensing image of a region to be detected and establishing a high-definition remote sensing data atlas of the region to be detected; extracting color and contour features of the region to be detected and establishing a standard feature map dataset; removing the impact of haze on the quality of remote sensing images based on a dark channel prior dehazing algorithm; segmenting the acquired remote sensing images based on the color and contour features of the region to be detected; calculating the blurriness of each portion of the remote sensing image and judging the quality of each portion of the remote sensing image; replacing and reconstructing poor-quality portions of the remote sensing image based on a neural network model; and optimizing the replaced and reconstructed remote sensing images based on a Laplace pyramid algorithm. The present invention has the advantages of effectively improving the output quality of remote sensing images and reducing the impact of haze and blur or occlusion of certain regions.

Description

Technical Field

[0001] The present invention relates to the field of image segmentation and reconstruction, and in particular to a multi-scale image segmentation and reconstruction method and system applied to remote sensing images. Background Art

[0002] Remote sensing images are image data of ground or surface objects collected through remote sensing technology. Remote sensing images can be used to monitor environmental changes, resource distribution and disaster warnings in the monitored area in real time. Ensuring the quality of remote sensing images and improving their accuracy and reliability are of great significance. Since the collection of remote sensing images is affected by the environment at the time, haze or cloudy environments can easily lead to a decline in the quality of remote sensing images, which is not conducive to the analysis of the detection area through remote sensing images. Therefore, it is of great significance to improve the quality of remote sensing images, remove the influence of haze through image segmentation and reconstruction, and improve the clarity and accuracy of remote sensing images.

[0003] Existing image segmentation and reconstruction techniques combine a dark channel prior algorithm with a U-Net remote sensing image to improve remote sensing image quality, reducing the impact of fog. For example, Chinese patent CN113763488A discloses a "remote sensing image dehazing method combining a dark channel prior algorithm and a U-Net." This technique reconstructs one channel of a remote sensing image and combines it with the other two channels to obtain three new haze images. It also uses a deep learning U-Net network to segment out dense fog areas in the remote sensing image and estimate atmospheric light values. Three transmission maps are then generated based on the three new haze images. Finally, an atmospheric scattering model is used to dehaze the three new haze images. The final dehazing effect is achieved by weighting the images. Although this type of technique can reduce the impact of fog on remote sensing images, it cannot always avoid its effects. Even in harsh environments, this technique struggles to achieve the desired effect. Most importantly, it struggles to avoid obstructions caused by other obstacles, resulting in localized shadows and poor final remote sensing image quality. Summary of the Invention

[0004] In order to solve the above technical problems, a multi-scale image segmentation and reconstruction method and system for remote sensing images are provided. This technical solution solves the problem that the influence of fog cannot be avoided in the above background technology. Even in harsh environments, this technology is difficult to achieve the expected effect. Most importantly, it is difficult to avoid occlusion by other obstacles, resulting in local occlusion shadows, causing the final remote sensing image quality to be poor.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0006] A multi-scale image segmentation and reconstruction method applied to remote sensing images, comprising:

[0007] Based on the remote sensing image database, obtain high-definition remote sensing images of the area to be inspected and establish a high-definition remote sensing data atlas of the inspection area;

[0008] According to the type of event in the area to be detected, the color and contour features of the area to be detected are extracted, and a standard feature map dataset is established;

[0009] Based on the high-definition remote sensing data atlas of the detection area, the dark channel prior defogging algorithm is used to remove the impact of haze on the quality of remote sensing images;

[0010] According to the high-definition remote sensing data atlas of the detection area, the collected remote sensing images are segmented based on the color and contour features of the area to be detected;

[0011] Obtain the segmented remote sensing image, calculate the fuzziness of each part of the remote sensing image, and judge the quality of each part of the remote sensing image based on the fuzziness of each part of the remote sensing image;

[0012] According to the evaluation results of the quality of each part of the remote sensing image, the poor quality parts of the remote sensing image are replaced and reconstructed based on the neural network model;

[0013] Based on the Laplace pyramid algorithm, the remote sensing images after replacement and reconstruction are optimized.

[0014] Preferably, the obtaining of high-definition remote sensing images of the area to be detected based on the remote sensing image database and the establishment of a high-definition remote sensing data atlas of the detection area specifically include:

[0015] Based on the remote sensing image database, high-definition remote sensing images of the area to be inspected are obtained. The high-definition remote sensing images of the area to be inspected mainly include: all-round image information of the area to be inspected and color changes of the four seasons;

[0016] Organize, classify and number the high-definition remote sensing images of the inspection area, and establish a high-definition remote sensing data atlas of the inspection area.

[0017] Preferably, extracting the color and contour features of the area to be detected based on the type of event in the area to be detected and establishing a standard feature map data set specifically includes:

[0018] Extracting color and contour features of the area to be detected based on the type of events in the area to be detected, wherein the types of events include: buildings, plants, rocks, rivers, crops, terrain, and man-made objects;

[0019] According to the color and contour features of the area to be detected, based on the high-definition remote sensing data atlas of the detection area, the color and contour feature images of the area to be detected that meet the standards are screened out to establish a standard feature map dataset.

[0020] Preferably, removing the influence of haze on the quality of remote sensing images based on a dark channel prior defogging algorithm according to a high-definition remote sensing data atlas of the detection area specifically includes:

[0021] According to the high-definition remote sensing data atlas of the detection area, grayscale processing is performed on the high-definition remote sensing data atlas of the detection area to obtain a grayscale high-definition remote sensing data atlas of the detection area;

[0022] According to the grayscale high-definition remote sensing data atlas of the detection area, the quality of the high-definition remote sensing data atlas of the detection area is optimized based on the Gaussian filtering algorithm;

[0023] According to the optimized high-definition remote sensing data atlas of the detection area, a dark channel atlas of the optimized high-definition remote sensing data atlas of the detection area is obtained;

[0024] The dark channel image is a grayscale image of the same size as the original image obtained by calculating the minimum value of the three RGB color channels in the local area of ​​each pixel in the image. This grayscale image is the dark channel image.

[0025] According to the optimized dark channel atlas of the high-definition remote sensing data atlas of the detection area, the atmospheric light value of the dark channel prior defogging algorithm is obtained;

[0026] According to the atmospheric scattering coefficient and the depth of field of the scene light, the transmission rate of light through fog at each pixel in the dark channel atlas of the high-definition remote sensing data atlas of the detection area is obtained;

[0027] Based on the dark channel prior dehazing algorithm, an optimization model for the high-definition remote sensing data atlas of the detection area is established to remove the impact of haze on the quality of remote sensing images;

[0028] The dark channel priori defogging algorithm expression is:

[0029] Where, The foggy image is observed in Rank The pixel values ​​after Gaussian filtering, The haze-free image to be restored is Rank The pixel value of the column, is the transmission rate of light through fog, is the atmospheric light value, is the atmospheric scattering coefficient, is the depth of field of the scene light.

[0030] Preferably, the segmenting of the collected remote sensing image based on the color and contour features of the area to be detected according to the high-definition remote sensing data atlas of the detection area specifically includes:

[0031] According to the high-definition remote sensing data atlas of the detection area, different locations in the high-definition remote sensing data atlas of the detection area are marked and numbered according to the color and contour characteristics of the area to be detected;

[0032] Based on the high-definition remote sensing data atlas of the marked and numbered detection area, the remote sensing images collected in real time are segmented according to the positions of the marks and numbers, and the color and contour features of each segmented area are confirmed to be the most obvious;

[0033] According to the standard feature map dataset, the atlas after the standard feature map dataset is segmented is set as the target standard atlas.

[0034] Preferably, the acquiring of the segmented remote sensing image, calculating the blurriness of each part of the remote sensing image, and judging the quality of each part of the remote sensing image by the blurriness of each part of the remote sensing image specifically includes:

[0035] According to the results of image segmentation of the remote sensing image collected in real time according to the positions of the marks and numbers, the pixel values ​​of the segmented real-time remote sensing image are extracted;

[0036] According to the segmentation results of the target standard atlas, the optimized pixel values ​​of the segmented standard remote sensing image are extracted;

[0037] Through the normalization formula, the pixel values ​​after optimization of the real-time remote sensing image and the pixel values ​​after optimization of the standard remote sensing image are subjected to dimension-eliminating optimization processing;

[0038] Calculate the blurriness of each part of the remote sensing image according to the optimized pixel value of the processed real-time remote sensing image and the optimized pixel value of the standard remote sensing image;

[0039] According to the fuzziness of each part of the remote sensing image, a threshold is set to judge the quality of each part of the remote sensing image;

[0040] Determine whether the blurriness of the remote sensing image is greater than a threshold. If so, it indicates that the quality of the current part of the remote sensing image is poor and does not meet expectations. If not, it indicates that the quality of the current part of the remote sensing image meets expectations.

[0041] The fuzziness expression of each part of the remote sensing image is:

[0042] Where, is the fuzziness of each part of the remote sensing image, is the correlation factor, After segmentation, the real-time remote sensing image is optimized. Rank The pixel value of the column, After segmentation, the standard remote sensing image is optimized. Rank The pixel value of the column, is the number of pixel values ​​after segmentation and optimization of the real-time remote sensing image, The number of images in the target standard atlas that have the same pixel value at the same position.

[0043] Preferably, the replacing and reconstructing the poor quality parts of the remote sensing image based on the neural network model according to the evaluation results of the quality of each part of the remote sensing image specifically includes:

[0044] According to the target standard atlas, remote sensing images of the area to be detected in different states are obtained, and target atlas and training atlas are established respectively;

[0045] The training atlas and the evaluation results of each part of the remote sensing image are used as the input matrix, and the target atlas is used as the verification output matrix to construct the input and output matrix of the neural network;

[0046] Based on the neural network model, the input and output matrix data are trained and learned. The quality of each part of the newly received real-time remote sensing image is evaluated based on the model training results.

[0047] Determine whether the quality of each part of the newly received real-time remote sensing image meets the requirements. If so, it means that the quality of this part of the image meets the expectations, and the original image of this part is output. If not, it means that the quality of this part of the image does not meet the expectations, and the image of this part is replaced and reconstructed.

[0048] Preferably, the optimization processing of the replaced and reconstructed remote sensing image based on the Laplace pyramid algorithm specifically includes:

[0049] Based on the reconstruction results of the remote sensing image by the neural network model, the image pyramid model is constructed by sampling and processing the result image, separating the maximum image and the minimum image;

[0050] According to the image pyramid model, Gaussian blur is performed on the image of the next layer, and even rows and columns of the blurred image are deleted. This process is repeated to obtain a Gaussian pyramid model.

[0051] According to the Gaussian pyramid model and the Laplace pyramid algorithm, a remote sensing image reconstruction optimization model is constructed to optimize the remote sensing images after replacement and reconstruction.

[0052] The Laplace pyramid algorithm is:

[0053] Where, The Laplace pyramid The image data of the layer, Gaussian pyramid The image data of the layer, To convert the first The pixel values ​​are mapped to the target image The position of the upward sampling process, is the convolution symbol, for Gaussian kernel.

[0054] Furthermore, this solution proposes a multi-scale image segmentation and reconstruction system for remote sensing images, which is used to implement the multi-scale image segmentation and reconstruction method for remote sensing images as described above, including:

[0055] A data processing module is used to obtain high-definition remote sensing images of the area to be detected based on the remote sensing image database and establish a high-definition remote sensing data atlas of the detection area; extract the color and contour features of the area to be detected according to the type of event in the area to be detected, and establish a standard feature map data set;

[0056] A defogging optimization module is used to remove the influence of haze on the quality of remote sensing images based on a dark channel prior defogging algorithm according to a high-definition remote sensing data atlas of the detection area;

[0057] An image segmentation module is used to segment the collected remote sensing image based on the color and contour features of the area to be detected according to the high-definition remote sensing data atlas of the detection area;

[0058] The image reconstruction and optimization module is used to obtain the segmented remote sensing image, calculate the blur of each part of the remote sensing image, and judge the quality of each part of the remote sensing image based on the blur of each part of the remote sensing image; based on the judgment results of the quality of each part of the remote sensing image, based on the neural network model, replace and reconstruct the parts of the remote sensing image with poor quality; and optimize the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm.

[0059] Preferably, the data processing module specifically includes:

[0060] A high-definition remote sensing atlas unit, which is used to obtain high-definition remote sensing images of the area to be detected based on a remote sensing image database and establish a high-definition remote sensing data atlas of the detection area;

[0061] A feature extraction unit, configured to extract color and contour features of the area to be detected based on the type of event in the area to be detected, and to establish a standard feature map data set;

[0062] The image reconstruction and optimization module specifically includes:

[0063] A fuzziness unit is used to obtain the segmented remote sensing image, calculate the fuzziness of each part of the remote sensing image, and judge the quality of each part of the remote sensing image according to the fuzziness of each part of the remote sensing image;

[0064] A quality evaluation unit, configured to replace and reconstruct poor-quality parts of the remote sensing image based on a neural network model according to evaluation results of the quality of each part of the remote sensing image;

[0065] The reconstruction optimization unit is used to optimize the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] By obtaining high-definition remote sensing data maps of the detection area and extracting the color and contour features of the area to be detected, a high-definition remote sensing data atlas and a standard feature map dataset of the detection area are established. Based on this, the dark channel prior dehazing algorithm is used to remove the haze from the remote sensing images collected in real time, thereby improving the quality of the remote sensing images. Secondly, the blurriness of each part of the remote sensing image is determined by the optimized pixel values ​​of the real-time remote sensing image and the optimized pixel values ​​of the standard remote sensing image. According to the blurriness of each part of the remote sensing image, a threshold is set to judge the quality of each part of the remote sensing image. Furthermore, according to the evaluation results of the quality of each part of the remote sensing image, the quality of each part of the newly received real-time remote sensing image is evaluated based on the neural network model, so that the parts of the remote sensing image with poor quality are replaced and reconstructed. Finally, based on the Laplace pyramid algorithm, a remote sensing image reconstruction optimization model is constructed to optimize the replaced and reconstructed remote sensing images, effectively fusion the images of each segmented area, thereby effectively improving the output quality of the remote sensing image, and maximally reducing the influence of haze and partial regional blur or occlusion, ensuring the reliability and accuracy of the remote sensing image. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a flow chart of a multi-scale image segmentation and reconstruction method applied to remote sensing images of the present invention;

[0069] Figure 2 This is a flow chart of the present invention for removing the influence of haze on remote sensing image quality based on a dark channel prior defogging algorithm according to a high-definition remote sensing data atlas of the detection area;

[0070] Figure 3 A flow chart of obtaining a segmented remote sensing image, calculating the fuzziness of each part of the remote sensing image, and judging the quality of each part of the remote sensing image by the fuzziness of each part of the remote sensing image;

[0071] Figure 4This is a flowchart of optimizing the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm of the present invention. DETAILED DESCRIPTION

[0072] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0073] Reference Figure 1 As shown, a multi-scale image segmentation and reconstruction method applied to remote sensing images includes:

[0074] Based on the remote sensing image database, obtain high-definition remote sensing images of the area to be inspected and establish a high-definition remote sensing data atlas of the inspection area;

[0075] According to the type of event in the area to be detected, the color and contour features of the area to be detected are extracted, and a standard feature map dataset is established;

[0076] Based on the high-definition remote sensing data atlas of the detection area, the dark channel prior defogging algorithm is used to remove the impact of haze on the quality of remote sensing images;

[0077] According to the high-definition remote sensing data atlas of the detection area, the collected remote sensing images are segmented based on the color and contour features of the area to be detected;

[0078] Obtain the segmented remote sensing image, calculate the fuzziness of each part of the remote sensing image, and judge the quality of each part of the remote sensing image based on the fuzziness of each part of the remote sensing image;

[0079] According to the evaluation results of the quality of each part of the remote sensing image, the poor quality parts of the remote sensing image are replaced and reconstructed based on the neural network model;

[0080] Based on the Laplace pyramid algorithm, the remote sensing images after replacement and reconstruction are optimized.

[0081] It can be explained that this scheme obtains high-definition remote sensing data maps of the detection area and extracts the color and contour features of the area to be detected, establishes high-definition remote sensing data atlases and standard feature map datasets of the detection area, and on this basis, based on the dark channel prior defogging algorithm, removes haze from the remote sensing pictures collected in real time, thereby improving the quality of remote sensing images. Secondly, the blurriness of each part of the remote sensing image is determined through the optimized pixel values ​​of the real-time remote sensing image and the optimized pixel values ​​of the standard remote sensing image, and according to the blurriness of each part of the remote sensing image, a threshold is set to judge the quality of each part of the remote sensing image. Furthermore, according to the evaluation results of the quality of each part of the remote sensing image, the new remote sensing image is classified into the following categories: The quality of each part of the received real-time remote sensing image is evaluated, so as to replace and reconstruct the parts of the remote sensing image with poor quality. Finally, based on the Laplace pyramid algorithm, a remote sensing image reconstruction optimization model is constructed to optimize the remote sensing image after replacement and reconstruction, and effectively fuse the images of each segmented area, thereby effectively improving the output quality of the remote sensing image, and reducing the influence of haze and blur or occlusion of some areas with the greatest efficiency, so as to ensure the reliability and accuracy of the remote sensing image. Among them, the remote sensing image database obtains the remote sensing image data of the detection area through satellite remote sensing, aerial remote sensing and UAV remote sensing, and records and saves the historical remote sensing image data of the detection area.

[0082] Reference Figure 2 As shown, the effects of haze on remote sensing image quality are removed based on the dark channel prior defogging algorithm according to the high-definition remote sensing data atlas of the detection area, specifically including:

[0083] According to the high-definition remote sensing data atlas of the detection area, grayscale processing is performed on the high-definition remote sensing data atlas of the detection area to obtain a grayscale high-definition remote sensing data atlas of the detection area;

[0084] According to the grayscale high-definition remote sensing data atlas of the detection area, the quality of the high-definition remote sensing data atlas of the detection area is optimized based on the Gaussian filtering algorithm;

[0085] According to the optimized high-definition remote sensing data atlas of the detection area, a dark channel atlas of the optimized high-definition remote sensing data atlas of the detection area is obtained;

[0086] The dark channel image is a grayscale image of the same size as the original image obtained by calculating the minimum value of the three RGB color channels in the local area of ​​each pixel in the image. This grayscale image is the dark channel image.

[0087] According to the optimized dark channel atlas of the high-definition remote sensing data atlas of the detection area, the atmospheric light value of the dark channel prior defogging algorithm is obtained;

[0088] According to the atmospheric scattering coefficient and the depth of field of the scene light, the transmission rate of light through fog at each pixel in the dark channel atlas of the high-definition remote sensing data atlas of the detection area is obtained;

[0089] Based on the dark channel prior dehazing algorithm, an optimization model for the high-definition remote sensing data atlas of the detection area is established to remove the impact of haze on the quality of remote sensing images;

[0090] The dark channel priori defogging algorithm expression is:

[0091] Where, The foggy image is observed in Rank The pixel values ​​after Gaussian filtering, The haze-free image to be restored is Rank The pixel value of the column, is the transmission rate of light through fog, is the atmospheric light value, is the atmospheric scattering coefficient, is the depth of field of the scene light.

[0092] It can be explained that the dark channel prior dehazing algorithm is a classic and effective image dehazing algorithm in the field of computer vision. The algorithm points out that in most natural images, there is at least one color channel whose pixel value is very close to zero in at least one local window. This channel, called the "dark channel", can be used to estimate the global atmospheric light in the image, and by analyzing the relationship between the distance between pixel points and the atmospheric light intensity, the haze information in the image can be inferred. Finally, this information is used to eliminate the haze in the image. This scheme introduces the dark channel prior dehazing algorithm to reduce the impact of the degradation of remote sensing image quality caused by haze or cloudy weather. By establishing an optimization model for the high-definition remote sensing data atlas of the detection area, the impact of haze is removed, thereby improving the quality of remote sensing images and facilitating subsequent analysis.

[0093] Reference Figure 3 As shown, the steps of obtaining the segmented remote sensing image, calculating the blurriness of each part of the remote sensing image, and judging the quality of each part of the remote sensing image by the blurriness of each part of the remote sensing image specifically include:

[0094] According to the results of image segmentation of the remote sensing image collected in real time according to the positions of the marks and numbers, the pixel values ​​of the segmented real-time remote sensing image are extracted;

[0095] According to the segmentation results of the target standard atlas, the pixel values ​​of the optimized standard remote sensing image after segmentation are extracted;

[0096] Through the normalization formula, the pixel values ​​after optimization of the real-time remote sensing image and the pixel values ​​after optimization of the standard remote sensing image are subjected to dimension-eliminating optimization processing;

[0097] Calculate the blurriness of each part of the remote sensing image according to the optimized pixel value of the processed real-time remote sensing image and the optimized pixel value of the standard remote sensing image;

[0098] According to the fuzziness of each part of the remote sensing image, a threshold is set to judge the quality of each part of the remote sensing image;

[0099] Determine whether the blurriness of the remote sensing image is greater than a threshold. If so, it indicates that the quality of the current part of the remote sensing image is poor and does not meet expectations. If not, it indicates that the quality of the current part of the remote sensing image meets expectations.

[0100] The fuzziness expression of each part of the remote sensing image is:

[0101] Where, is the fuzziness of each part of the remote sensing image, is the correlation factor, After segmentation, the real-time remote sensing image is optimized. Rank The pixel value of the column, After segmentation, the standard remote sensing image is optimized. Rank The pixel value of the column, is the number of pixel values ​​after segmentation and optimization of the real-time remote sensing image, The number of images in the target standard atlas that have the same pixel value at the same position.

[0102] It can be explained that when judging the quality of remote sensing images, it is necessary to consider the image quality of each segmented area of ​​the remote sensing image in order to facilitate subsequent reconstruction. This scheme determines the fuzziness of each part of the remote sensing image by establishing the fuzziness of each part of the remote sensing image and analyzing the relationship between the pixel values ​​of the optimized real-time remote sensing image after segmentation and the pixel values ​​of the optimized standard remote sensing image after segmentation. Among them, the greater the fuzziness, the worse the image quality of the segmented area, and the greater the possibility of replacement.

[0103] Reference Figure 4 As shown, the optimization process of the replaced and reconstructed remote sensing image based on the Laplace pyramid algorithm specifically includes:

[0104] Based on the reconstruction results of the remote sensing image by the neural network model, the image pyramid model is constructed by sampling and processing the result image, separating the maximum image and the minimum image;

[0105] According to the image pyramid model, Gaussian blur is performed on the image of the next layer, and even rows and columns of the blurred image are deleted. This process is repeated to obtain a Gaussian pyramid model.

[0106] According to the Gaussian pyramid model and the Laplace pyramid algorithm, a remote sensing image reconstruction optimization model is constructed to optimize the remote sensing images after replacement and reconstruction.

[0107] The Laplace pyramid algorithm is:

[0108] Where, The Laplace pyramid The image data of the layer, Gaussian pyramid The image data of the layer, To convert the first The pixel values ​​are mapped to the target image The position of the upward sampling process, is the convolution symbol, for Gaussian kernel.

[0109] It can be explained that the remote sensing images are replaced and reconstructed through the neural network model, and the optimal stitchable images are selected for stitching and reconstruction according to the color and contour characteristics of the area to be detected. However, it is difficult for the neural network model to effectively complete the perfect fusion of remote sensing images. When processing images, boundary blur or dislocation problems may occur, resulting in overall disharmony of the remote sensing images. This scheme introduces the Laplace pyramid algorithm. Its multi-scale decomposition and reconstruction characteristics enable it to better fuse image information from different modalities. The algorithm can effectively fuse the images of each segmented area to ensure the overall quality and smoothness of the remote sensing image, thereby further improving the quality of remote sensing images.

[0110] Furthermore, based on the same inventive concept as the multi-scale image segmentation and reconstruction method applied to remote sensing images, this solution proposes a multi-scale image segmentation and reconstruction system applied to remote sensing images, comprising:

[0111] A data processing module is used to obtain high-definition remote sensing images of the area to be detected based on the remote sensing image database and establish a high-definition remote sensing data atlas of the detection area; extract the color and contour features of the area to be detected according to the type of event in the area to be detected, and establish a standard feature map data set;

[0112] A defogging optimization module is used to remove the influence of haze on the quality of remote sensing images based on a dark channel prior defogging algorithm according to a high-definition remote sensing data atlas of the detection area;

[0113] An image segmentation module is used to segment the collected remote sensing image based on the color and contour features of the area to be detected according to the high-definition remote sensing data atlas of the detection area;

[0114] An image reconstruction and optimization module is used to obtain the segmented remote sensing image, calculate the blur of each part of the remote sensing image, and judge the quality of each part of the remote sensing image based on the blur of each part of the remote sensing image; based on the judgment results of the quality of each part of the remote sensing image, replace and reconstruct the parts of the remote sensing image with poor quality based on the neural network model; and optimize the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm;

[0115] The data processing module includes:

[0116] A high-definition remote sensing atlas unit, which is used to obtain high-definition remote sensing images of the area to be detected based on a remote sensing image database and establish a high-definition remote sensing data atlas of the detection area;

[0117] A feature extraction unit, configured to extract color and contour features of the area to be detected based on the type of event in the area to be detected, and to establish a standard feature map data set;

[0118] The image reconstruction and optimization module includes:

[0119] A fuzziness unit is used to obtain the segmented remote sensing image, calculate the fuzziness of each part of the remote sensing image, and judge the quality of each part of the remote sensing image according to the fuzziness of each part of the remote sensing image;

[0120] A quality evaluation unit, configured to replace and reconstruct poor-quality parts of the remote sensing image based on a neural network model according to evaluation results of the quality of each part of the remote sensing image;

[0121] The reconstruction optimization unit is used to optimize the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm.

[0122] In summary, the advantages of the present invention are: effectively fusing the images of each segmented area, thereby effectively improving the output quality of the remote sensing image, reducing the impact of haze and blur or occlusion of some areas with the greatest efficiency, and ensuring the reliability and accuracy of the remote sensing image.

[0123] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-scale image segmentation and reconstruction method for remote sensing images, characterized in that: include: Based on the remote sensing image database, obtain high-definition remote sensing images of the area to be inspected and establish a high-definition remote sensing data atlas of the inspection area; According to the type of event in the area to be detected, the color and contour features of the area to be detected are extracted, and a standard feature map dataset is established; Based on the high-definition remote sensing data atlas of the detection area, the dark channel prior defogging algorithm is used to remove the impact of haze on the quality of remote sensing images; According to the high-definition remote sensing data atlas of the detection area, the collected remote sensing images and the standard feature map dataset are segmented based on the color and contour features of the area to be detected, and the atlas after the segmentation of the standard feature map dataset is set as the target standard atlas; Obtain the segmented remote sensing image, calculate the fuzziness of each part of the remote sensing image, and judge the quality of each part of the remote sensing image based on the fuzziness of each part of the remote sensing image; According to the evaluation results of the quality of each part of the remote sensing image, the poor quality parts of the remote sensing image are replaced and reconstructed based on the neural network model; Based on the Laplace pyramid algorithm, the remote sensing images after replacement and reconstruction are optimized; The replacing and reconstructing the poor quality parts of the remote sensing image based on the neural network model according to the evaluation results of the quality of each part of the remote sensing image specifically includes: According to the target standard atlas, remote sensing images of the area to be detected in different states are obtained, and target atlas and training atlas are established respectively; The training atlas and the evaluation results of each part of the remote sensing image are used as the input matrix, and the target atlas is used as the verification output matrix to construct the input and output matrix of the neural network; Based on the neural network model, the input and output matrix data are trained and learned. The quality of each part of the newly received real-time remote sensing image is evaluated based on the model training results. Determine whether the quality of each part of the newly received real-time remote sensing image meets the requirements. If so, it means that the quality of this part of the image meets the expectations, and the original image of this part is output. If not, it means that the quality of this part of the image does not meet the expectations, and the image of this part is replaced and reconstructed.

2. The multi-scale image segmentation and reconstruction method for remote sensing images according to claim 1, characterized in that: The method of obtaining high-definition remote sensing images of the area to be detected based on the remote sensing image database and establishing a high-definition remote sensing data atlas of the detection area specifically includes: Based on the remote sensing image database, high-definition remote sensing images of the area to be inspected are obtained. The high-definition remote sensing images of the area to be inspected mainly include: all-round image information of the area to be inspected and color changes of the four seasons; Organize, classify and number the high-definition remote sensing images of the inspection area, and establish a high-definition remote sensing data atlas of the inspection area.

3. The multi-scale image segmentation and reconstruction method for remote sensing images according to claim 2, characterized in that: The method of extracting the color and contour features of the area to be detected according to the type of the event in the area to be detected and establishing a standard feature map dataset specifically includes: Extracting color and contour features of the area to be detected based on the type of events in the area to be detected, wherein the types of events include: buildings, plants, rocks, rivers, crops, terrain, and man-made objects; According to the color and contour features of the area to be detected, based on the high-definition remote sensing data atlas of the detection area, the color and contour feature images of the area to be detected that meet the standards are screened out to establish a standard feature map dataset.

4. The multi-scale image segmentation and reconstruction method for remote sensing images according to claim 3, characterized in that: The above method, based on the high-definition remote sensing data atlas of the detection area and based on the dark channel prior defogging algorithm, removes the impact of haze on the quality of remote sensing images, specifically including: According to the high-definition remote sensing data atlas of the detection area, grayscale processing is performed on the high-definition remote sensing data atlas of the detection area to obtain a grayscale high-definition remote sensing data atlas of the detection area; According to the grayscale high-definition remote sensing data atlas of the detection area, the quality of the high-definition remote sensing data atlas of the detection area is optimized based on the Gaussian filtering algorithm; According to the optimized high-definition remote sensing data atlas of the detection area, a dark channel atlas of the optimized high-definition remote sensing data atlas of the detection area is obtained; The dark channel image is a grayscale image of the same size as the original image obtained by calculating the minimum value of the three RGB color channels in the local area of ​​each pixel in the image. This grayscale image is the dark channel image. According to the optimized dark channel atlas of the high-definition remote sensing data atlas of the detection area, the atmospheric light value of the dark channel prior defogging algorithm is obtained; According to the atmospheric scattering coefficient and the depth of field of the scene light, the transmission rate of light through fog at each pixel in the dark channel atlas of the high-definition remote sensing data atlas of the detection area is obtained; Based on the dark channel prior dehazing algorithm, an optimization model for the high-definition remote sensing data atlas of the detection area is established to remove the impact of haze on the quality of remote sensing images; The dark channel priori defogging algorithm expression is: Where, The foggy image is observed in Rank The pixel values ​​after Gaussian filtering, The haze-free image to be restored is Rank The pixel value of the column, is the transmission rate of light through fog, is the atmospheric light value, is the atmospheric scattering coefficient, is the depth of field of the scene light.

5. The multi-scale image segmentation and reconstruction method for remote sensing images according to claim 4, characterized in that: The method of segmenting the collected remote sensing image based on the color and contour features of the area to be detected according to the high-definition remote sensing data atlas of the detection area specifically includes: According to the high-definition remote sensing data atlas of the detection area, different locations in the high-definition remote sensing data atlas of the detection area are marked and numbered according to the color and contour characteristics of the area to be detected; Based on the high-definition remote sensing data atlas of the marked and numbered detection area, the remote sensing images collected in real time are segmented according to the positions of the marks and numbers, and the color and contour features of each segmented area are confirmed to be the most obvious; According to the standard feature map dataset, the atlas after the standard feature map dataset is segmented is set as the target standard atlas.

6. The multi-scale image segmentation and reconstruction method for remote sensing images according to claim 5, characterized in that: The step of obtaining the segmented remote sensing image, calculating the blurriness of each portion of the remote sensing image, and judging the quality of each portion of the remote sensing image based on the blurriness of each portion of the remote sensing image specifically includes: According to the results of image segmentation of the remote sensing image collected in real time according to the positions of the marks and numbers, the pixel values ​​of the segmented real-time remote sensing image are extracted; According to the segmentation results of the target standard atlas, the optimized pixel values ​​of the segmented standard remote sensing image are extracted; Through the normalization formula, the pixel values ​​after optimization of the real-time remote sensing image and the pixel values ​​after optimization of the standard remote sensing image are subjected to dimension-eliminating optimization processing; Calculate the blurriness of each part of the remote sensing image according to the optimized pixel value of the processed real-time remote sensing image and the optimized pixel value of the standard remote sensing image; According to the fuzziness of each part of the remote sensing image, a threshold is set to judge the quality of each part of the remote sensing image; Determine whether the blurriness of the remote sensing image is greater than a threshold. If so, it indicates that the quality of the current part of the remote sensing image is poor and does not meet expectations. If not, it indicates that the quality of the current part of the remote sensing image meets expectations. The fuzziness expression of each part of the remote sensing image is: Where, is the fuzziness of each part of the remote sensing image, is the correlation factor, After segmentation, the real-time remote sensing image is optimized. Rank The pixel value of the column, After segmentation, the standard remote sensing image is optimized. Rank The pixel value of the column, is the number of pixel values ​​after segmentation and optimization of the real-time remote sensing image, The number of images in the target standard atlas that have the same pixel value at the same position.

7. The multi-scale image segmentation and reconstruction method for remote sensing images according to claim 6, characterized in that: The optimization process of the replaced and reconstructed remote sensing image based on the Laplace pyramid algorithm specifically includes: Based on the reconstruction results of the remote sensing image by the neural network model, the image pyramid model is constructed by sampling and processing the result image, separating the maximum image and the minimum image; According to the image pyramid model, Gaussian blur is performed on the image of the next layer, and even rows and columns of the blurred image are deleted. This process is repeated to obtain a Gaussian pyramid model. According to the Gaussian pyramid model and the Laplace pyramid algorithm, a remote sensing image reconstruction optimization model is constructed to optimize the remote sensing images after replacement and reconstruction. The Laplace pyramid algorithm is: Where, The Laplace pyramid The image data of the layer, Gaussian pyramid The image data of the layer, To convert the first The pixel values ​​are mapped to the target image The position of the upward sampling process, is the convolution symbol, for Gaussian kernel.

8. A multi-scale image segmentation and reconstruction system for remote sensing images, characterized in that: A method for implementing a multi-scale image segmentation and reconstruction method for remote sensing images according to any one of claims 1 to 7, comprising: A data processing module is used to obtain high-definition remote sensing images of the area to be detected based on the remote sensing image database and establish a high-definition remote sensing data atlas of the detection area; extract the color and contour features of the area to be detected according to the type of event in the area to be detected, and establish a standard feature map data set; A defogging optimization module is used to remove the influence of haze on the quality of remote sensing images based on a dark channel prior defogging algorithm according to a high-definition remote sensing data atlas of the detection area; An image segmentation module is used to segment the collected remote sensing image based on the color and contour features of the area to be detected according to the high-definition remote sensing data atlas of the detection area; The image reconstruction and optimization module is used to obtain the segmented remote sensing image, calculate the blur of each part of the remote sensing image, and judge the quality of each part of the remote sensing image based on the blur of each part of the remote sensing image; based on the judgment results of the quality of each part of the remote sensing image, based on the neural network model, replace and reconstruct the parts of the remote sensing image with poor quality; and optimize the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm.

9. The multi-scale image segmentation and reconstruction system for remote sensing images according to claim 8, characterized in that: The data processing module specifically includes: A high-definition remote sensing atlas unit, which is used to obtain high-definition remote sensing images of the area to be detected based on a remote sensing image database and establish a high-definition remote sensing data atlas of the detection area; A feature extraction unit, configured to extract color and contour features of the area to be detected based on the type of event in the area to be detected, and to establish a standard feature map data set; The image reconstruction and optimization module specifically includes: A fuzziness unit is used to obtain the segmented remote sensing image, calculate the fuzziness of each part of the remote sensing image, and judge the quality of each part of the remote sensing image according to the fuzziness of each part of the remote sensing image; A quality evaluation unit, configured to replace and reconstruct poor-quality parts of the remote sensing image based on a neural network model according to evaluation results of the quality of each part of the remote sensing image; The reconstruction optimization unit is used to optimize the remote sensing image after replacement and reconstruction based on the Laplace pyramid algorithm.

Images