High-precision Remote Sensing Image Data Fusion Method and System for Power Transmission and Transformation Ring Water and Soil Conservation

Abstract

The present invention provides a method and a system for high-precision remote sensing image data fusion for the environmental protection of the water circulation in power transmission and transformation. The method first performs upsampling on the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and performs image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image; then, based on the first feature extraction model, the spectral features of the hyperspectral remote sensing image are extracted to obtain remote sensing spectral features; based on the second feature extraction model, the texture features of the high-resolution panchromatic image are extracted to obtain remote sensing texture features; finally, the remote sensing spectral features, the remote sensing texture features and the high-resolution panchromatic image are fused to obtain a high-resolution multispectral image. The present invention completes the fusion of high-resolution images and low-resolution images in the process of satellite remote sensing monitoring of the environmental protection of the water circulation in power transmission and transformation projects to generate high-resolution multispectral images, assisting the environmental protection monitoring personnel for water circulation to better complete the environmental monitoring tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a high-precision remote sensing image data fusion method and system for the water and soil conservation of power transmission and transformation projects. Background Art

[0002] Remote sensing technology is a technology that remotely senses and measures the Earth's surface through satellites or other high-altitude platforms. With the development of technology, remote sensing technology has been widely applied in water environment monitoring, providing powerful technical support for the protection and utilization of water resources. However, due to the differences in shooting equipment and shooting parameters, remote sensing images with different resolutions will appear. For remote sensing images with different resolutions, they have different advantages in water and soil conservation monitoring. For example, high-resolution panchromatic images are rich in spatial structure information and can express the detailed features of ground objects in detail. Low-resolution multispectral images are rich in spectral information and are beneficial to the identification and interpretation of ground objects. Therefore, how to generate a high-resolution multispectral image that combines the advantages of both has become one of the problems in the satellite remote sensing monitoring of water and soil conservation for power transmission and transformation projects. Therefore, the present invention proposes a high-precision remote sensing image data fusion method and system for the water and soil conservation of power transmission and transformation projects. Summary of the Invention

[0003] The present invention provides a high-precision remote sensing image data fusion method and system for the water and soil conservation of power transmission and transformation projects, which completes the fusion of high-resolution images and low-resolution images during the satellite remote sensing monitoring process of the water and soil conservation of power transmission and transformation projects to generate high-resolution multispectral images, and assists the water and soil conservation monitoring personnel to better complete the environmental monitoring tasks.

[0004] The present invention provides a high-precision remote sensing image data fusion method for the water and soil conservation of power transmission and transformation projects, including:

[0005] Step 1: Upsample the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and perform image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image;

[0006] Step 2: Extract the spectral features of the hyperspectral remote sensing image based on the first feature extraction model to obtain remote sensing spectral features;

[0007] Step 3: Extract the texture features of the high-resolution panchromatic image based on the second feature extraction model to obtain remote sensing texture features;

[0008] Step 4: Fuse the remote sensing spectral features, remote sensing texture features and the high-resolution panchromatic image to obtain a high-resolution multispectral image.

[0009] Preferably, in a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection of water circulation, image enhancement is performed on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image, including:

[0010] According to the preset standard ground object distribution characteristics of the power transmission and transformation project, the initial high-resolution remote sensing image is identified to determine the areas of the power transmission and transformation project station area and the areas along the line.

[0011] Based on the areas of the power transmission and transformation project station area and the areas along the line, as well as the preset protection range, and in combination with the scale of the remote sensing image, the effective water circulation and environmental protection areas are determined on the initial low-resolution remote sensing image and its corresponding initial high-resolution remote sensing image with the same spatial resolution, respectively.

[0012] Taking the effective water circulation and environmental protection area as the image enhancement center, based on the attention mechanism algorithm, image enhancement is performed on the initial high-resolution remote sensing image to obtain a high-resolution panchromatic image.

[0013] Preferably, in a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection of water circulation, both the first feature extraction model and the second feature extraction model are multi-scale feature extraction models, and their feature extraction scales are the same.

[0014] Preferably, in a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection of water circulation, it further includes:

[0015] The determination of the feature extraction scales of the first feature extraction model and the second feature extraction model, the specific method includes:

[0016] Compare the resolutions of the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution.

[0017] And synchronously obtain the standard feature extraction size density in the data storage module.

[0018] Based on the minimum resolution, in combination with the standard feature extraction size density, obtain the feature extraction scales corresponding to the first feature extraction model and

[0019] the second feature extraction model.

[0020] Preferably, in a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection of water circulation, it further includes:

[0021] Obtain multiple historical high-resolution remote sensing images, and based on multiple random feature extraction size densities, in combination with the initial resolution and the maximum scaling scale of the historical high-resolution remote sensing images, generate multiple first feature extraction scales.

[0022] Based on multiple first feature extraction scales, layer separation is performed on multiple high-resolution remote sensing images respectively to obtain a multi-first historical layer pyramid, and the layer data between the historical layers of the same first historical layer pyramid is compared;

[0023] Based on the comparison results, the first data redundancy rate of each first historical layer pyramid is determined, and the data redundancy rates of the first historical layer pyramids corresponding to the same random feature extraction size density are weighted and averaged to obtain the first average redundancy rate;

[0024] Based on the first average redundancy rate, the optimal feature extraction size density is determined, and according to the adjacent feature extraction size densities corresponding to the optimal feature extraction size density, the optimal extraction density range is determined;

[0025] The optimal extraction density range is evenly divided into three small ranges: upper, middle, and lower. If the optimal feature extraction size density is within the middle small range, then the optimal feature extraction size density is used as the final feature extraction size density;

[0026] Otherwise, the middle feature extraction size densities of the upper, middle, and lower small ranges are respectively obtained as the test feature extraction size densities. Based on the test feature extraction size densities, combined with the initial resolution and the maximum zoom scale of the historical high-resolution remote sensing images, multiple second feature extraction scales are generated;

[0027] Based on the second feature extraction scales, layer separation is performed on multiple high-resolution remote sensing images respectively to obtain a second historical layer pyramid, and the layer data between the historical layers of the same second historical layer pyramid is compared;

[0028] Based on the comparison results, the second data redundancy rate of each second historical layer pyramid is determined, and it is judged and the data redundancy rates of the second historical layer pyramids corresponding to the same test feature extraction size density are weighted and averaged to obtain the second average redundancy rate;

[0029] Judge whether there is a value in the second average redundancy rate that is less than the first average redundancy rate corresponding to the optimal feature extraction scale density;

[0030] If not, then the optimal feature extraction size density is used as the final feature extraction size density;

[0031] Otherwise, the optimal feature extraction size density corresponding to the minimum second average redundancy rate is used as the final feature extraction size density;

[0032] The final feature extraction size density is stored as the standard feature extraction size density.

[0033] Preferably, in a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection of water circulation, fusing remote sensing spectral features, remote sensing texture features, and high-resolution panchromatic images to obtain a high-resolution multispectral image includes:

[0034] Obtain the feature extraction scales of the first feature extraction model and the second feature extraction model, and process the high-resolution panchromatic image based on the feature extraction scales to obtain an image pyramid;

[0035] Based on the resolution corresponding to each layer of the image pyramid, obtain the corresponding remote sensing sub-spectral features and remote sensing sub-texture features, and perform same-layer feature fusion on the current layer with the remote sensing sub-spectral features and remote sensing sub-texture features to obtain a fused layer;

[0036] Based on the upsampling method, fuse the top-layer fused layer and the low-layer fused layer to obtain a high-resolution multispectral image.

[0037] Preferably, in a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection of water circulation, after obtaining the high-resolution multispectral image, it further includes:

[0038] Detect the quality of the high-resolution multispectral image, and determine whether the clarity of the effective water circulation and environmental protection area in the high-resolution multispectral image reaches a preset standard. If so, perform edge enhancement processing on the effective water circulation and environmental protection area in the high-resolution multispectral image to obtain the final output image;

[0039] Otherwise, perform denoising processing on the high-resolution multispectral image and then extract feature points, and obtain the information entropy difference between any two neighborhood information entropies of each feature point. When the information entropy difference is greater than a preset value, determine that the feature point is an unstable point;

[0040] Use the neighborhood corresponding to the larger entropy value among the any two neighborhood information entropies as the target neighborhood, and based on the target neighborhood and the pixel position of the unstable point on the high-resolution multispectral image, perform supplementary area positioning on the initial high-resolution remote sensing image;

[0041] Perform interpolation processing on the unstable point based on the supplementary neighborhood, and perform edge enhancement on the effective water circulation and environmental protection area in the new high-resolution multispectral image obtained after the interpolation processing to obtain the final output image.

[0042] A high-precision remote sensing image data fusion system for power transmission and transformation environmental protection of water circulation according to the present invention includes:

[0043] An initial image processing module, configured to perform upsampling on an initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and perform image enhancement on the initial high-resolution remote sensing image corresponding to the initial low-resolution remote sensing image with the same spatial resolution to obtain a high-resolution panchromatic image;

[0044] A spectral feature extraction module, which is used to extract the spectral features of hyperspectral remote sensing images based on a first feature extraction model to obtain remote sensing spectral features;

[0045] A texture feature extraction module, which is used to extract the texture features of high-resolution panchromatic images based on a second feature extraction model to obtain remote sensing texture features;

[0046] A remote sensing image fusion module, which is used to fuse the remote sensing spectral features, remote sensing texture features and high-resolution panchromatic images to obtain a high-resolution multispectral image.

[0047] Preferably, in a high-precision remote sensing image data fusion system for the environmental protection of the water circulation in the power transmission and transformation loop, it further includes: a feature extraction preparation module, which is used to determine the feature extraction scales of the first feature extraction model and the second feature extraction model, including:

[0048] A comparison processing unit, which is used to compare the corresponding resolutions of the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution;

[0049] A data acquisition unit, which is used to acquire the standard feature extraction size density in the data storage module;

[0050] A scale determination unit, which is used to obtain the feature extraction scales corresponding to the first feature extraction model and the second feature extraction model based on the minimum resolution and in combination with the standard feature extraction size density.

[0051] Preferably, in a high-precision remote sensing image data fusion system for the environmental protection of the water circulation in the power transmission and transformation loop, it further includes: an output processing module, which is used for:

[0052] Detect the quality of the high-resolution multispectral image, and judge whether the clarity of the effective water circulation and environmental protection area in the high-resolution multispectral image reaches a preset standard. If so, perform edge enhancement processing on the effective water circulation and environmental protection area in the high-resolution multispectral image to obtain a final output image;

[0053] Otherwise, perform denoising processing on the high-resolution multispectral image and then perform feature point extraction, and obtain the information entropy difference between any two neighborhood information entropies of each feature point. When the information entropy difference is greater than a preset value, determine that the feature point is an unstable point;

[0054] Use the neighborhood corresponding to the larger entropy value among the any two neighborhood information entropies as the target neighborhood, and based on the target neighborhood and the pixel position of the unstable point on the high-resolution multispectral image, perform supplementary area positioning on the initial high-resolution remote sensing image;

[0055] Interpolate the unstable points based on the supplementary area, and perform edge enhancement on the effective water and soil conservation area in the new high-resolution multispectral image obtained after interpolation to obtain the final output image.

[0056] Compared with the prior art, the present invention has the following beneficial effects: By performing upsampling on the initial low-resolution remote sensing image, a hyperspectral remote sensing image is obtained, and image enhancement is performed on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image; then, based on the first feature extraction model, the spectral features of the hyperspectral remote sensing image are extracted to obtain remote sensing spectral features; based on the second feature extraction model, the texture features of the high-resolution panchromatic image are extracted to obtain remote sensing texture features, realizing the targeted acquisition of the advantageous features of low-resolution and high-resolution images; finally, the remote sensing spectral features, remote sensing texture features, and high-resolution panchromatic image are fused to obtain a high-resolution multispectral image. Only the advantageous features are obtained for fusion, effectively reducing the interference data in the fusion process and facilitating the improvement of the accuracy of the spectral features and texture features presented in the high-resolution multispectral image. The present invention completes the fusion of high-resolution and low-resolution images in the process of satellite remote sensing monitoring of water and soil conservation in power transmission and transformation projects to generate high-resolution multispectral images, assisting water and soil conservation monitoring personnel to better complete environmental monitoring tasks.

[0057] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in this application document.

[0058] The technical solutions of the present invention will be further described in detail below with reference to the drawings and embodiments. Description of the Drawings

[0059] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0060] Figure 1 is a flowchart of a high-precision remote sensing image data fusion method for water and soil conservation in power transmission and transformation of the present invention;

[0061] Figure 2 is a flowchart of step 4 of a high-precision remote sensing image data fusion method for water and soil conservation in power transmission and transformation of the present invention;

[0062] Figure 3 is a schematic diagram of a high-precision remote sensing image data fusion system for water and soil conservation in power transmission and transformation of the present invention. Detailed Embodiments

[0063] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0064] Embodiment 1:

[0065] The present invention provides a method for fusing high-precision remote sensing image data for the environmental protection of power transmission and transformation loop water, as Figure 1 shown, including:

[0066] Step 1: Upsample the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and perform image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image;

[0067] Step 2: Extract the spectral features of the hyperspectral remote sensing image based on the first feature extraction model to obtain remote sensing spectral features;

[0068] Step 3: Extract the texture features of the high-resolution panchromatic image based on the second feature extraction model to obtain remote sensing texture features;

[0069] Step 4: Fuse the remote sensing spectral features, remote sensing texture features and the high-resolution panchromatic image to obtain a high-resolution multispectral image;

[0070] Among them, both the first feature extraction model and the second feature extraction model are multi-scale feature extraction models, and their feature extraction scales are the same;

[0071] The multi-scale feature extraction models are all deep convolutional neural networks.

[0072] The beneficial effects of the above technical solutions: The present invention upsamples the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and performs image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image; then extracts the spectral features of the hyperspectral remote sensing image based on the first feature extraction model to obtain remote sensing spectral features; extracts the texture features of the high-resolution panchromatic image based on the second feature extraction model to obtain remote sensing texture features, realizing the targeted acquisition of the advantageous features of low-resolution images and high-resolution images; finally, fuses the remote sensing spectral features, remote sensing texture features and the high-resolution panchromatic image to obtain a high-resolution multispectral image, only fusing the advantageous features, effectively reducing the interference data in the fusion process, and being beneficial to improving the accuracy of the spectral features and texture features presented by the high-resolution multispectral image. The present invention completes the fusion of high-resolution images and low-resolution images in the satellite remote sensing monitoring process of the environmental protection of power transmission and transformation projects to generate high-resolution multispectral images, assisting environmental protection monitoring personnel to better complete environmental monitoring tasks.

[0073] Embodiment 2:

[0074] Based on Embodiment 1, perform image enhancement on the initial high-resolution remote sensing image corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image, including:

[0075] Identify the initial high-resolution remote sensing image according to the preset standard ground object distribution characteristics of the power transmission and transformation project, and determine the areas of the power transmission and transformation project station area and the areas along the line;

[0076] Based on the areas of the power transmission and transformation project station area and the areas along the line and the preset protection range, combined with the scale of the remote sensing image, determine the effective water and soil conservation areas on the initial low-resolution remote sensing image and its corresponding initial high-resolution remote sensing image with the same spatial resolution respectively;

[0077] Use the effective water and soil conservation area as the image enhancement center, and perform image enhancement on the initial high-resolution remote sensing image based on the attention mechanism algorithm to obtain a high-resolution panchromatic image.

[0078] In this embodiment, the standard ground object distribution characteristics refer to the characteristics of relevant typical ground objects of the power transmission and transformation project (for example, cables, power transmission towers, power transmission and transformation project station areas, etc.).

[0079] In this embodiment, the preset protection range refers to the protection range of the water area and the environment near the power transmission and transformation project preset by the user. For example, it is a circle with a radius of 10 kilometers near the power transmission and transformation project station area.

[0080] Advantages of the above technical solution: According to the preset standard ground object distribution characteristics of the power transmission and transformation project, the present invention identifies the initial high-resolution remote sensing image to determine the areas of the power transmission and transformation project station area and the areas along the line; based on the areas of the power transmission and transformation project station area and the areas along the line and the preset protection range, combined with the scale of the remote sensing image, determine the effective water and soil conservation areas on the initial low-resolution remote sensing image and its corresponding initial high-resolution remote sensing image with the same spatial resolution respectively; use the effective water and soil conservation area as the image enhancement center, and perform image enhancement on the initial high-resolution remote sensing image based on the attention mechanism algorithm to obtain a high-resolution panchromatic image, precisely enhance the effective water and soil conservation area on the remote sensing image, make the detail features in the effective water and soil conservation area more prominent, which is beneficial to improving the accuracy of the extracted remote sensing texture features and reducing the loss rate of high-frequency signals in the effective water and soil conservation area during the fusion process.

[0081] Embodiment 3:

[0082] Based on Embodiment 1, a high-precision remote sensing image data fusion method for power transmission and transformation water and soil conservation further includes:

[0083] Determine the feature extraction scales of the first feature extraction model and the second feature extraction model. The specific method includes:

[0084] Compare the resolutions of the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution;

[0085] And synchronously obtain the standard feature extraction size density in the data storage module;

[0086] Based on the minimum resolution, combined with the standard feature extraction size density, obtain the feature extraction scales corresponding to the first feature extraction model and the second feature extraction model.

[0087] In this embodiment, the feature extraction scale refers to the number of layers included in the image pyramid during the multi-scale feature extraction process.

[0088] Beneficial effects of the above technical solution: The present invention first compares the resolutions of the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution; and synchronously obtains the standard feature extraction size density in the data storage module; subsequently, based on the minimum resolution, combined with the standard feature extraction size density, obtain the feature extraction scales corresponding to the first feature extraction model and the second feature extraction model, realizing the customization of the multi-scale feature extraction scale of the remote sensing image, and can reduce the data redundancy rate during the multi-scale feature fusion process.

[0089] Embodiment 4:

[0090] Based on Embodiment 3, a high-precision remote sensing image data fusion method for the water and soil conservation of power transmission and transformation also includes:

[0091] Obtain multiple historical high-resolution remote sensing images, and based on multiple random feature extraction size densities, combined with the initial resolution and the maximum scaling scale of the historical high-resolution remote sensing images, generate multiple first feature extraction scales;

[0092] Based on the multiple first feature extraction scales, separate the layers of multiple high-resolution remote sensing images respectively to obtain multiple first historical layer pyramids, and compare the layer data between the historical layers of the same first historical layer pyramid;

[0093] Based on the comparison results, determine the first data redundancy rate of each first historical layer pyramid, and perform weighted averaging on the data redundancy rates of the first historical layer pyramids corresponding to the same random feature extraction size density to obtain the first average redundancy rate;

[0094] Based on the first average redundancy rate, determine the optimal feature extraction size density, and according to the adjacent feature extraction size densities corresponding to the optimal feature extraction size density, determine the optimal extraction density range;

[0095] Evenly divide the range of the optimal extraction density to obtain three small ranges: upper, middle, and lower. If the optimal feature extraction size density is within the middle small range, then use the optimal feature extraction size density as the final feature extraction size density;

[0096] Otherwise, respectively obtain the intermediate feature extraction size densities of the upper, middle, and lower small ranges as the test feature extraction size densities. Based on the test feature extraction size densities, combined with the initial resolution and the maximum zoom scale of the historical high-resolution remote sensing images, generate multiple second feature extraction scales;

[0097] Based on the second feature extraction scales, perform layer separation on multiple high-resolution remote sensing images respectively to obtain the second historical layer pyramid, and compare the layer data between the historical layers of the same second historical layer pyramid;

[0098] Based on the comparison results, determine the second data redundancy rate of each second historical layer pyramid, judge and perform weighted averaging on the data redundancy rates of the second historical layer pyramids corresponding to the same test feature extraction size density to obtain the second average redundancy rate;

[0099] Judge whether there is a value in the second average redundancy rate that is less than the first average redundancy rate corresponding to the optimal feature extraction scale density;

[0100] If not, then use the optimal feature extraction size density as the final feature extraction size density;

[0101] Otherwise, use the optimal feature extraction size density corresponding to the minimum second average redundancy rate as the final feature extraction size density;

[0102] Store the final feature extraction size density as the standard feature extraction size density.

[0103] In this embodiment, the feature extraction size density refers to the resolution difference between the layers during the multi-scale feature extraction process.

[0104] In this embodiment, the first feature extraction scale refers to the number of layers obtained by allocating the layer resolution according to the initial resolution and the maximum zoom scale according to the random feature extraction size density.

[0105] In this embodiment, the first data redundancy rate refers to the average value of the proportion of redundant data in the comparison of all layers in each first historical layer pyramid or second historical layer pyramid with adjacent layers. When a certain layer has two adjacent layers, respectively obtain the proportion of redundant data between the layer and the upper and lower adjacent layers and perform a comparison, and use the larger proportion of redundant data as the data redundancy rate of the layer.

[0106] In this embodiment, the first average redundancy rate refers to the average value of the data redundancy rates of all the first historical layer pyramids corresponding to the same random feature extraction size density of all the historical high-resolution remote sensing images.

[0107] In this embodiment, the adjacent feature extraction size density refers to one or two feature extraction size densities adjacent to the optimal feature extraction size density among all the random feature extraction size densities.

[0108] In this embodiment, the optimal extraction density range refers to the extraction density value range with the adjacent feature extraction size densities as the upper and lower limits. When there is only one adjacent feature extraction size density for the optimal feature extraction size density, the optimal feature extraction size density is the upper or lower limit value of the optimal extraction density range.

[0109] In this embodiment, the intermediate feature extraction size density refers to the middle value of the upper, middle, and lower ranges. The test feature extraction size density is a test value used to detect whether the optimal feature extraction size density is indeed the optimal feature extraction size density.

[0110] In this embodiment, the second feature extraction scale refers to the number of layers obtained by allocating layer resolution according to the test feature extraction size density based on the initial resolution and the maximum zoom scale.

[0111] In this embodiment, the second average redundancy rate refers to the average value of the data redundancy rates of all the second historical layer pyramids corresponding to the same test feature extraction size density of all the historical high-resolution remote sensing images.

[0112] Beneficial effects of the above technical solution: The present invention randomly generates multiple feature extraction size densities, separates layers of multiple historical high-resolution remote sensing images according to the randomly generated multiple feature extraction size densities, then determines the first data redundancy rate of each first historical layer pyramid based on the obtained historical layer pyramid, and performs weighted averaging on the data redundancy rates of the first historical layer pyramids corresponding to the same random feature extraction size density to obtain the first average redundancy rate; based on the first average redundancy rate, determines the optimal feature extraction size density, determines the optimal extraction density range according to the adjacent feature extraction size densities corresponding to the optimal feature extraction size density, and then evenly divides the optimal extraction density range into three small ranges, namely upper, middle, and lower. If the optimal feature extraction size density is within the middle small range, then the optimal feature extraction size density is used as the final feature extraction size density. Otherwise, the middle feature extraction size densities of the upper, middle, and lower small ranges are respectively obtained as the test feature extraction size densities. Based on the test feature extraction size densities, combined with the initial resolution and the maximum scaling factor of the historical high-resolution remote sensing image, multiple second feature extraction scales are generated; based on the second feature extraction scales, layers of multiple high-resolution remote sensing images are separated respectively to obtain a second historical layer pyramid, and the layer data between the respective historical layers of the same second historical layer pyramid are compared; based on the comparison result, determines the second data redundancy rate of each second historical layer pyramid, judges and performs weighted averaging on the data redundancy rates of the second historical layer pyramids corresponding to the same test feature extraction size density to obtain the second average redundancy rate; determines whether there is a value in the second average redundancy rate that is less than the first average redundancy rate corresponding to the optimal feature extraction scale density; if not, then the optimal feature extraction size density is used as the final feature extraction size density; otherwise, the optimal feature extraction size density corresponding to the minimum second average redundancy rate is used as the final feature extraction size density; and stores the final feature extraction size density as the standard feature extraction size density, realizing the adaptive customization of the number of layers for multi-scale feature extraction of remote sensing layers, while ensuring the accuracy of feature extraction, minimizing the data redundancy rate in the layer separation process.

[0113] Example 5:

[0114] On the basis of Example 1, step 4: Fuse the remote sensing spectral features, remote sensing texture features, and high-resolution panchromatic image to obtain a high-resolution multispectral image, as Figure 2 shown, including:

[0115] Step 401: Obtain the feature extraction scales of the first feature extraction model and the second feature extraction model, and process the high-resolution panchromatic image based on the feature extraction scales to obtain an image pyramid;

[0116] Step 402: Based on the resolution corresponding to each layer of the image pyramid, obtain the corresponding remote sensing sub-spectral features and remote sensing sub-texture features, and perform same-layer feature fusion on the current layer with the remote sensing sub-spectral features and remote sensing sub-texture features to obtain a fused layer;

[0117] Step 403: Based on the upsampling method, fuse the top-layer fused layer and the low-layer fused layer to obtain a high-resolution multispectral image.

[0118] Beneficial effects of the above technical solution: The present invention first takes the feature extraction scales of the first feature extraction model and the second feature extraction model, processes the high-resolution panchromatic image based on the feature extraction scales to obtain an image pyramid, and then based on the resolution corresponding to each layer of the image pyramid, obtains the corresponding remote sensing sub-spectral features and remote sensing sub-texture features, and performs same-layer feature fusion on the current layer with the remote sensing sub-spectral features and remote sensing sub-texture features to obtain a fused layer, realizing the fusion of the image features of a single layer of the image pyramid. Finally, based on the upsampling method, the top-layer fused layer and the low-layer fused layer are fused to obtain a high-resolution multispectral image, realizing the fusion of images with different resolutions in the power transmission and transformation project.

[0119] Embodiment 6:

[0120] On the basis of Embodiment 5, after obtaining the high-resolution multispectral image, it further includes:

[0121] Detect the quality of the high-resolution multispectral image, and determine whether the clarity of the effective environmental protection area with water and soil conservation in the high-resolution multispectral image reaches the preset standard. If so, after performing edge enhancement processing on the effective environmental protection area with water and soil conservation in the high-resolution multispectral image, obtain the final output image;

[0122] Otherwise, perform denoising processing on the high-resolution multispectral image and then perform feature point extraction, and obtain the information entropy difference between any two neighborhood information entropies of each feature point. When the information entropy difference is greater than the preset value, determine that the feature point is an unstable point;

[0123] Take the neighborhood corresponding to the larger entropy value in the any two neighborhood information entropies as the target neighborhood, and based on the target neighborhood and the pixel position of the unstable point on the high-resolution multispectral image, perform supplementary area positioning on the initial high-resolution remote sensing image;

[0124] Perform interpolation processing on the unstable point based on the supplementary neighborhood, and perform edge enhancement on the effective environmental protection area with water and soil conservation in the new high-resolution multispectral image obtained after the interpolation processing to obtain the final output image.

[0125] Beneficial effects of the above technical solution: After obtaining a high-resolution multispectral image, the present invention detects the quality of the high-resolution multispectral image and determines whether the clarity of the effective environmental protection area with circular water in the high-resolution multispectral image reaches a preset standard. If so, after performing edge enhancement processing on the effective environmental protection area with circular water in the high-resolution multispectral image, a final output image is obtained; otherwise, after denoising the high-resolution multispectral image, feature points are extracted, and the information entropy difference between any two neighborhood information entropies of each feature point is obtained. When the information entropy difference is greater than a preset value, the feature point is determined to be an unstable point; the neighborhood corresponding to the larger entropy value among the any two neighborhood information entropies is used as the target neighborhood, and based on the target neighborhood and the pixel position of the unstable point on the high-resolution multispectral image, supplementary area positioning is performed on the initial high-resolution remote sensing image; interpolation processing is performed on the unstable point based on the supplementary neighborhood, and edge enhancement is performed on the effective environmental protection area with circular water in the newly obtained high-resolution multispectral image after interpolation processing to obtain a final output image, overcoming the situation that multi-scale feature fusion may cause blurring of some regions of the image, instability of some pixel points, and weakening of high-frequency signals. Before output, edge enhancement is performed on the effective environmental protection area with circular water in the high-resolution multispectral image, realizing the compensation of high-frequency signals while enhancing and highlighting abnormal positions in the high-resolution remote sensing image, which is also beneficial for environmental protection monitoring personnel to timely discover pollution sources and abnormal positions of power supply.

[0126] Embodiment 7:

[0127] The present invention provides a high-precision remote sensing image data fusion system for environmental protection of power transmission and transformation, as Figure 3 shown, including:

[0128] An initial image processing module, configured to perform upsampling on an initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and perform image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image;

[0129] A spectral feature extraction module, configured to extract spectral features of the hyperspectral remote sensing image based on a first feature extraction model to obtain remote sensing spectral features;

[0130] A texture feature extraction module, configured to extract texture features of the high-resolution panchromatic image based on a second feature extraction model to obtain remote sensing texture features;

[0131] A remote sensing image fusion module, configured to fuse the remote sensing spectral features, remote sensing texture features, and the high-resolution panchromatic image to obtain a high-resolution multispectral image.

[0132] Advantages of the above technical solution: In the present invention, the initial image processing module performs upsampling on the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and performs image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image; then, through the spectral feature extraction module and the texture feature extraction module, based on the first feature extraction model, the spectral features of the hyperspectral remote sensing image are extracted to obtain remote sensing spectral features; based on the second feature extraction model, the texture features of the high-resolution panchromatic image are extracted to obtain remote sensing texture features, realizing the targeted acquisition of the dominant features of low-resolution images and high-resolution images; finally, the remote sensing image fusion module fuses the remote sensing spectral features, remote sensing texture features, and high-resolution panchromatic image to obtain a high-resolution multispectral image. Only the dominant features are acquired for fusion, effectively reducing the interference data in the fusion process, and facilitating the improvement of the accuracy of the spectral features and texture features presented in the high-resolution multispectral image. The present invention completes the fusion of high-resolution images and low-resolution images in the process of satellite remote sensing monitoring of environmental protection and water conservation for power transmission and transformation projects to generate high-resolution multispectral images, assisting environmental protection and water conservation monitoring personnel to better complete environmental monitoring tasks.

[0133] Embodiment 8:

[0134] Based on Embodiment 7, a high-precision remote sensing image data fusion system for environmental protection and water conservation of power transmission and transformation further includes: a feature extraction preparation module for determining the feature extraction scales of the first feature extraction model and the second feature extraction model, including:

[0135] A comparison processing unit for comparing the resolutions corresponding to the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution;

[0136] A data acquisition unit for acquiring the standard feature extraction size density in the data storage module;

[0137] A scale determination unit for obtaining the feature extraction scales corresponding to the first feature extraction model and the second feature extraction model based on the minimum resolution and in combination with the standard feature extraction size density.

[0138] Embodiment 9:

[0139] Based on Embodiment 7, a high-precision remote sensing image data fusion system for environmental protection and water conservation of power transmission and transformation further includes: an output processing module for:

[0140] Detecting the quality of the high-resolution multispectral image and determining whether the clarity of the effective environmental protection and water conservation area in the high-resolution multispectral image reaches a preset standard. If so, after performing edge enhancement processing on the effective environmental protection and water conservation area in the high-resolution multispectral image, a final output image is obtained;

[0141] Otherwise, perform denoising processing on the high-resolution multispectral image, then extract feature points, and obtain the information entropy difference between any two neighborhood information entropies of each feature point. When the information entropy difference is greater than a preset value, determine that the feature point is an unstable point;

[0142] Take the neighborhood corresponding to the larger entropy value among the any two neighborhood information entropies as the target neighborhood. Based on the target neighborhood and the pixel position of the unstable point on the high-resolution multispectral image, perform supplementary area positioning on the initial high-resolution remote sensing image;

[0143] Perform interpolation processing on the unstable point based on the supplementary neighborhood, and perform edge enhancement on the effective water and soil conservation area in the new high-resolution multispectral image obtained after the interpolation processing to obtain the final output image.

[0144] Beneficial effects of the above technical solution: After obtaining the high-resolution multispectral image, the present invention detects the quality of the high-resolution multispectral image, and determines whether the clarity of the effective water and soil conservation area in the high-resolution multispectral image reaches a preset standard. If so, perform edge enhancement processing on the effective water and soil conservation area in the high-resolution multispectral image to obtain the final output image; otherwise, perform denoising processing on the high-resolution multispectral image, then extract feature points, and obtain the information entropy difference between any two neighborhood information entropies of each feature point. When the information entropy difference is greater than a preset value, determine that the feature point is an unstable point; take the neighborhood corresponding to the larger entropy value among the any two neighborhood information entropies as the target neighborhood. Based on the target neighborhood and the pixel position of the unstable point on the high-resolution multispectral image, perform supplementary area positioning on the initial high-resolution remote sensing image; perform interpolation processing on the unstable point based on the supplementary neighborhood, and perform edge enhancement on the effective water and soil conservation area in the new high-resolution multispectral image obtained after the interpolation processing to obtain the final output image. It overcomes the situation that multi-scale feature fusion will cause partial blurring of the image, instability of some pixel points, and weakening of high-frequency signals. Before output, perform edge enhancement on the effective water and soil conservation area in the high-resolution multispectral image, realizing the compensation of high-frequency signals while enhancing and highlighting the abnormal positions in the high-resolution remote sensing image, which is also beneficial for water and soil conservation monitoring personnel to timely discover pollution sources and abnormal positions of power supply.

[0145] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A high-precision remote sensing image data fusion method for power transmission and transformation environmental protection and water conservation, characterized in that: include: Step 1: Upsample the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and perform image enhancement on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image; Step 2: extracting the spectral features of the hyperspectral remote sensing image based on the first feature extraction model to obtain remote sensing spectral features; Step 3: Extract texture features of the high-resolution full-color image based on the second feature extraction model to obtain remote sensing texture features; Step 4: Fuse the remote sensing spectral features, remote sensing texture features and high-resolution panchromatic images to obtain high-resolution multispectral images; Wherein, the first feature extraction model and the second feature extraction model are both multi-scale feature extraction models, and the feature extraction scales of the two are consistent; The feature extraction scales of the first feature extraction model and the second feature extraction model are determined, and the specific method includes: Compare the resolutions of the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution; and simultaneously obtain the standard feature extraction size density in the data storage module; Based on the minimum resolution and combined with the standard feature extraction size density, the feature extraction scales corresponding to the first feature extraction model and the second feature extraction model are obtained.

2. According to claim 1, a high-precision remote sensing image data fusion method for power transmission and transformation environmental protection and water conservation is characterized in that: Image enhancement is performed on the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution full-color image, including: According to the preset standard ground feature distribution characteristics of the power transmission and transformation project, the initial high-resolution remote sensing image is identified to determine the power transmission and transformation project station area and the area along the line; Based on the transmission and transformation project station area, the area along the line and the preset protection range, combined with the scale of the remote sensing image, the effective environmental and water conservation area is determined on the initial low-resolution remote sensing image and its corresponding initial high-resolution remote sensing image with the same spatial resolution; Taking the effective environmental and water conservation area as the image enhancement center, the initial high-resolution remote sensing image is enhanced based on the attention mechanism algorithm to obtain a high-resolution full-color image.

3. The high-precision remote sensing image data fusion method for power transmission and transformation environmental protection and water conservation according to claim 1 is characterized in that: Also includes: Acquire a plurality of historical high-resolution remote sensing images, and generate a plurality of first feature extraction scales based on a plurality of random feature extraction size densities combined with an initial resolution and a maximum zoom scale of the historical high-resolution remote sensing images; Based on multiple first feature extraction scales, multiple high-resolution remote sensing images are layer separated to obtain multiple first historical layer pyramids, and layer data between various historical layers of the same first historical layer pyramid are compared; Based on the comparison result, determine the first data redundancy rate of each first historical layer pyramid, and perform weighted average on the data redundancy rates of the first historical layer pyramids corresponding to the same random feature extraction size density to obtain a first average redundancy rate; Based on the first average redundancy rate, determining the optimal feature extraction size density, and determining the optimal extraction density range according to the adjacent feature extraction size densities corresponding to the optimal feature extraction size density; The optimal extraction density range is divided into intervals to obtain three small ranges of upper, middle and lower. If the optimal feature extraction size density is in the middle small range, the optimal feature extraction size density is used as the final feature extraction size density. Otherwise, the intermediate feature extraction size densities of the upper, middle and lower small ranges are respectively obtained as the test feature extraction size density, and multiple second feature extraction scales are generated based on the test feature extraction size density, combined with the initial resolution and maximum zoom scale of the historical high-resolution remote sensing image; Based on the second feature extraction scale, multiple high-resolution remote sensing images are layer separated to obtain a second historical layer pyramid, and layer data between various historical layers of the same second historical layer pyramid are compared; Based on the comparison result, determine the second data redundancy rate of each second historical layer pyramid, judge and perform weighted average of the data redundancy rates of the second historical layer pyramids corresponding to the same inspection feature extraction size density to obtain a second average redundancy rate; Determine whether there is a value in the second average redundancy rate that is smaller than the first average redundancy rate corresponding to the optimal feature extraction scale density; If it does not exist, then the best feature extraction size density is used as the final feature extraction size density; Otherwise, the optimal feature extraction size density corresponding to the minimum second average redundancy rate is used as the final feature extraction size density; The final feature extraction size density is stored as the standard feature extraction size density.

4. The high-precision remote sensing image data fusion method for power transmission and transformation environmental protection and water conservation according to claim 1 is characterized in that: The remote sensing spectral features, remote sensing texture features and high-resolution panchromatic images are integrated to obtain high-resolution multispectral images, including: Acquire feature extraction scales of the first feature extraction model and the second feature extraction model, and perform in-and-out processing on the high-resolution full-color image based on the feature extraction scales to obtain an image pyramid; Based on the resolution corresponding to each layer of the image pyramid, the corresponding remote sensing sub-spectral features and remote sensing sub-texture features are obtained, and the current layer is fused with the remote sensing sub-spectral features and remote sensing sub-texture features to obtain a fused layer; Based on the upsampling method, the top fusion layer is fused with the lower fusion layer to obtain a high-resolution multispectral image.

5. The high-precision remote sensing image data fusion method for power transmission and transformation environmental protection and water conservation according to claim 4 is characterized in that: After obtaining high-resolution multispectral images, it also includes: Detect the quality of the high-resolution multispectral image and determine whether the clarity of the effective environmental and water conservation area in the high-resolution multispectral image meets the preset standard. If so, perform edge enhancement processing on the effective environmental and water conservation area in the high-resolution multispectral image to obtain the final output image; Otherwise, the high-resolution multispectral image is subjected to denoising and then feature point extraction is performed, and the information entropy difference between any two domain information entropies of each feature point is obtained. When the information entropy difference is greater than a preset value, the feature point is determined to be an unstable point. The area corresponding to the larger entropy value of the information entropy of any two areas is taken as the target area, and the complementary area positioning is performed on the initial high-resolution remote sensing image based on the target area and the pixel position of the unstable point on the high-resolution multispectral image; The unstable points are interpolated based on the supplementary fields, and the edges of the effective environmental and water conservation areas in the new high-resolution multispectral image obtained after the interpolation are enhanced to obtain the final output image.

6. A high-precision remote sensing image data fusion system for power transmission and transformation environmental and water conservation, characterized in that: include: The initial image processing module is used to upsample the initial low-resolution remote sensing image to obtain a hyperspectral remote sensing image, and to enhance the initial high-resolution remote sensing image with the same spatial resolution corresponding to the initial low-resolution remote sensing image to obtain a high-resolution panchromatic image; A spectral feature extraction module is used to extract spectral features of a hyperspectral remote sensing image based on a first feature extraction model to obtain remote sensing spectral features; A texture feature extraction module is used to extract texture features of a high-resolution full-color image based on a second feature extraction model to obtain remote sensing texture features; Remote sensing image fusion module, used to fuse remote sensing spectral features, remote sensing texture features and high-resolution panchromatic images to obtain high-resolution multispectral images; Wherein, the first feature extraction model and the second feature extraction model are both multi-scale feature extraction models, and the feature extraction scales of the two are consistent; The system further comprises: The feature extraction preparation module is used to determine the feature extraction scale of the first feature extraction model and the second feature extraction model, including: A contrast processing unit, used to compare the resolutions corresponding to the hyperspectral remote sensing image and the high-resolution panchromatic image to obtain the minimum resolution; A data acquisition unit, used for acquiring the standard feature extraction size density in the data storage module; The scale determination unit is used to obtain the feature extraction scales corresponding to the first feature extraction model and the second feature extraction model based on the minimum resolution and in combination with the standard feature extraction size density.

7. The high-precision remote sensing image data fusion system for power transmission and transformation environmental and water conservation according to claim 6 is characterized in that: Also includes: Output processing module for: Detect the quality of the high-resolution multispectral image and determine whether the clarity of the effective environmental and water conservation area in the high-resolution multispectral image meets the preset standard. If so, perform edge enhancement processing on the effective environmental and water conservation area in the high-resolution multispectral image to obtain the final output image; Otherwise, the high-resolution multispectral image is subjected to denoising and then feature point extraction is performed, and the information entropy difference between any two domain information entropies of each feature point is obtained. When the information entropy difference is greater than a preset value, the feature point is determined to be an unstable point. The area corresponding to the larger entropy value of the information entropy of any two areas is taken as the target area, and the complementary area positioning is performed on the initial high-resolution remote sensing image based on the target area and the pixel position of the unstable point on the high-resolution multispectral image; The unstable points are interpolated based on the supplementary fields, and the edges of the effective environmental and water conservation areas in the new high-resolution multispectral image obtained after the interpolation are enhanced to obtain the final output image.

Images