Terrain survey image enhancement method for modular substation arrangement

Abstract

The invention relates to the field of image enhancement, in particular to a topographic survey image enhancement method for modular substation placement, and the method comprises the steps: carrying out the convolution processing of a topographic image, and obtaining first feature maps of different scales; the size of a normalization window in an image local contrast normalization algorithm is adaptively adjusted, and the image local contrast normalization algorithm is utilized to carry out normalization processing on the first feature maps of all scales to obtain second feature maps; calculating a cross-scale fusion weight of each sub-region in the second feature map; and fusing the sub-regions at the same position by using the cross-scale fusion weight to obtain fusion regions, wherein a plurality of fusion regions form an enhanced topographic image. According to the method, feature details of different scales can be reserved, microscopic details are prevented from being too smooth or a macroscopic structure is prevented from being fuzzy, and the requirement of modular substation site selection can be met.

Description

Technical Field

[0001] The present invention relates to the field of image enhancement, and particularly to a method for enhancing terrain survey images for modular substation placement. Background Art

[0002] The site selection of modular substations highly depends on terrain conditions, and it is necessary to accurately analyze surface morphology, geological stability, and potential risks (such as landslides, faults). However, in actual survey images, due to terrain undulations, weather conditions, and equipment limitations, local overexposure or underexposure occurs in the images, resulting in the loss of details in shadow areas (such as fissures, rock layer textures) and the distortion of features in highlight areas (such as reflection artifacts). The terrain has both macroscopic structures (such as ridges, valleys) and microscopic details (such as rock fissures, soil particles).

[0003] A Chinese patent document with the publication number CN111612700B discloses an image enhancement method, which includes: determining the clarity ratio of pixel points in each color channel of an initial RGB image; converting the initial RGB image into at least one reference YUV image; fusing each reference YUV image according to the clarity ratio of pixel points in each color channel of the initial RGB image to obtain a target YUV image, and converting the target YUV image into a target RGB image; determining the target RGB image as the enhanced image of the initial RGB image.

[0004] The prior art enhances the initial image through traditional single-scale enhancement methods. However, this image enhancement method cannot retain the feature details of different scales, resulting in over-smoothing of microscopic details or blurring of macroscopic structures, and it cannot meet the needs of modular substation site selection. Therefore, how to retain the feature details of different scales during image enhancement processing is the problem to be solved by the present invention. Summary of the Invention

[0005] To solve the problem of retaining the feature details of different scales during image enhancement processing, the present invention provides a method for enhancing terrain survey images for modular substation placement.

[0006] The present invention provides a method for enhancing terrain survey images for modular substation placement, adopting the following technical solutions: Obtain a terrain image, and perform convolution processing on the terrain image to obtain first feature maps of different scales; Adaptively adjust the size of the normalization window in the image local contrast normalization algorithm, and use the image local contrast normalization algorithm to perform normalization processing on the first feature maps of each scale to obtain second feature maps; Divide each second feature map into multiple sub-regions, and for multiple second feature maps, the sub-regions at the same position correspond to each other; Calculate the cross-scale fusion weights for each sub-region in the second feature map. The cross-scale fusion weights are positively correlated with the information entropy of the sub-regions. Use the cross-scale fusion weights to fuse the sub-regions at the same position to obtain a fused region, and multiple fused regions constitute the enhanced topographic image.

[0007] By performing convolution processing on the topographic image to obtain multiple first feature maps, normalizing the first feature maps, and then using the cross-scale fusion weights to fuse the sub-regions to obtain the enhanced topographic image, it is possible to retain the feature details at different scales, avoid over-smoothing of micro details or blurring of macro structures, and meet the needs of modular substation site selection.

[0008] Preferably, the method further includes: calculating an adjustment coefficient for the normalized window size, and the expression is:

[0009] In the formula, represents the adjustment coefficient of the normalized window size of the local image contrast of pixel point , represents the gradient magnitude of pixel point , represents the standard deviation of the brightness values of the neighboring pixel points of pixel point , represents the maximum gradient magnitude of the pixel points in this scale feature map, represents the maximum value of the standard deviation of the brightness values of the neighboring pixel points of all pixel points in this scale feature map.

[0010] Calculating the adjustment coefficient through the above formula improves the accuracy of the calculation result of the adjustment coefficient and provides a theoretical basis for calculating the size of the normalized window. Preferably, the size of the normalized window in the adaptive local image contrast normalization algorithm is adjusted as:

[0011] In the formula, represents the adjusted normalized window of pixel point , represents the size of the preset window, represents the adjustment coefficient of the normalized window size of the local image contrast of pixel point , exp represents the exponential function with base e, represents the floor function symbol.

[0012] By adjusting the size of the normalized window, the normalized image can retain as many detail features as possible.

[0013] Preferably, the expression of the cross-scale fusion weight is:

[0014] In the formula, represents the cross-scale fusion weight of the sub-region in the p-th second feature map, represents the information entropy of the brightness value of the sub-region in the p-th second feature map, and p represents the index of the second feature map, represents the number of second feature maps.

[0015] Calculating the cross-scale fusion weight of the sub-region through the information entropy of the sub-region can avoid the details being covered during fusion, thereby retaining more details.

[0016] Preferably, before fusing the sub-regions at the same position using the cross-scale fusion weight, it further includes: calculating the thermal conductivity coefficient in the anisotropic diffusion filtering algorithm, and filtering the sub-regions in the second feature map using the anisotropic diffusion filtering algorithm.

[0017] Preferably, the expression of the thermal conductivity coefficient is:

[0018] In the formula, represents the self-adaptive adjusted thermal conductivity coefficient of the sub-region in the second feature map at scale , represents the preset thermal conductivity coefficient, represents the average value of the gradients of all pixel points of the sub-region in the second feature map at scale in the axis direction, represents the average value of the gradients of all pixel points of the sub-region in the second feature map at scale in the axis direction, represents the maximum value of the gradients in the sub-region.

[0019] By calculating the thermal conductivity coefficient of the sub-region, it can avoid excessive smoothing of details and suppress noise at the same time.

[0020] Preferably, the expression for fusing the sub-regions at the same position using the cross-scale fusion weight is:

[0021] In the formula, represents the cross-scale fusion brightness value at position in the fusion region, represents the number of second feature maps, represents the cross-scale fusion weight at position in the sub-region of the p-th second feature map, represents the position in the sub-region of the p-th second feature map The brightness value, where p represents the index of the second feature map.

[0022] Preferably, the method further includes filtering the enhanced topographic image using an anisotropic diffusion filtering algorithm.

[0023] Preferably, before performing convolution processing on the topographic image, it further includes: performing geometric correction and radiometric correction on the topographic image.

[0024] Preferably, before normalizing the first feature maps at each scale using the image local contrast normalization algorithm, it further includes: using bilinear interpolation upsampling for the first feature maps at different scales to make the size of the first feature maps the same as that of the topographic image.

[0025] The present invention has the following technical effects: By performing convolution processing on the topographic image to obtain multiple first feature maps, normalizing the first feature maps, and then fusing sub-regions using cross-scale fusion weights to obtain an enhanced topographic image, it is possible to retain feature details at different scales, avoid over-smoothing of micro details or blurring of macro structures, and be able to meet the needs of modular substation site selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flowchart of a method for enhancing topographic survey images for modular substation placement according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] An embodiment of the present invention discloses a method for enhancing topographic survey images for modular substation placement. Referring to Figure 1 , it includes the following steps, specifically as follows: S1: Obtain a topographic image, and perform convolution processing on the topographic image to obtain first feature maps at different scales.

[0029] A visible light image of the area to be surveyed is taken using a multispectral camera to obtain a topographic image. Geometric correction is performed on the topographic image to correct the deformation caused by factors such as photographic material deformation, objective lens distortion, atmospheric refraction, earth curvature, earth rotation, and terrain undulation, where the geometric positions, shapes, sizes, orientations, etc. of various features on the topographic image do not match the expression requirements in the reference system. Then, radiometric correction is performed on the topographic image to eliminate or correct the image brightness distortion caused by radiometric errors.

[0030] Three different-sized convolutional kernels of 3×3, 5×5, and 7×7 are respectively used to perform convolution on the corrected topographic image to obtain three first feature maps of different scales, which are used to capture the feature details of different levels in the area to be surveyed.

[0031] S2: Adaptively adjust the size of the normalization window in the image local contrast normalization algorithm, and use the image local contrast normalization algorithm to perform normalization processing on the first feature maps of each scale to obtain second feature maps.

[0032] Since the convolutional kernel sizes of the first feature maps of different scales are different, the first feature maps have different numerical ranges, and there are obvious differences in the statistical distribution in terms of mean and variance. For example, the overall brightness of the mountain contour is low and the average amplitude is low due to the influence of global shadows, while the brightness of fissures or gentle slopes is high and the average amplitude is high due to local reflection. Therefore, adaptively adjust the local contrast normalization window according to the gradient magnitude and regional contrast of pixel points, and perform normalization on the first feature maps of each scale through the adaptive normalization window.

[0033] Perform bilinear interpolation upsampling on the first feature maps of different scales to make the sizes of the first feature maps the same as that of the topographic image. Then, calculate the adjustment coefficient of the normalization window size in the image local contrast normalization algorithm, and the expression is:

[0034] In the formula, represents the adjustment coefficient of the image local contrast normalization window size of pixel point , represents the gradient magnitude of pixel point , represents the standard deviation of the brightness values of the neighboring pixel points of pixel point , represents the maximum gradient magnitude of pixel points in this scale feature map, represents the maximum value of the standard deviations of the brightness values of the neighboring pixel points of all pixel points in this scale feature map, where serves to normalize the adjustment coefficient.

[0035] Adaptively adjust the size of the normalization window in the image local contrast normalization algorithm. The method is:

[0036] In the formula, represents a pixel point the adjusted normalized window, represents the size of the preset window. Exemplarily, k = 3 indicates that the size of the window is 3×3. represents a pixel point is the adjustment coefficient of the size of the local contrast normalization window of the pixel point. exp represents the exponential function with base e. represents the floor symbol.

[0037] The greater the local contrast and the greater the gradient magnitude, it indicates that the pixel point may belong to areas with complex textures such as rock fractures and steep slopes, containing more details, and its normalized window should be smaller to retain details; the smaller the local contrast and the smaller the gradient magnitude, it indicates that the pixel point may belong to flat and single-texture areas such as sandy land, water areas, and homogeneous soil, with fewer details and stronger noise intensity, so its normalized window should be larger to suppress noise.

[0038] S3: Divide each second feature map into multiple sub-regions. For multiple second feature maps, the sub-regions at the same position correspond to each other, and calculate the cross-scale fusion weight of each sub-region in the second feature map. The fusion weight is positively correlated with the information entropy of the sub-region.

[0039] Each second feature map is divided into 9 sub-regions. It can be understood that the first sub-regions in the three second feature maps correspond to each other, the second sub-regions correspond to each other,..., and the ninth sub-regions correspond to each other.

[0040] The expression of the cross-scale fusion weight is:

[0041] In the formula, represents the cross-scale fusion weight of the sub-region in the p-th second feature map, represents the information entropy of the brightness value of the sub-region in the p-th second feature map. p represents the index of the second feature map. represents the number of second feature maps. It should be noted that each sub-region of the second feature map corresponds to a fusion weight. Exemplarily, if the information entropies of the second sub-regions in the three second feature maps are 1, 2, and 3 respectively, then the cross-scale fusion weights of the second sub-regions in the three second feature maps are , , .

[0042] The larger the information entropy, the more complex the key terrain feature details contained in the sub-regions of the second feature map at this scale. A higher cross-scale fusion weight needs to be given to retain the details and avoid the details being covered during fusion.

[0043] S4: Use the cross-scale fusion weights to fuse the sub-regions at the same position to obtain a fusion region, and multiple fusion regions constitute the enhanced terrain image.

[0044] In one embodiment, calculate the thermal conductivity coefficient in the anisotropic diffusion filtering algorithm, and use the anisotropic diffusion filtering algorithm to filter the sub-regions in the second feature map.

[0045] The expression of the thermal conductivity coefficient is:

[0046] In the formula, represents the adaptively adjusted thermal conductivity coefficient of the sub-region in the second feature map at scale , represents the preset thermal conductivity coefficient, The value of is set artificially according to the actual situation. Exemplarily, the value of is 15, represents the average value of the gradients of all pixel points in the sub-region in the second feature map at scale in the axis direction, represents the average value of the gradients of all pixel points in the sub-region in the second feature map at scale in the axis direction,

[0047] The larger the gradient magnitude of a region, the more texture details the region contains. A smaller thermal conductivity coefficient should be used to avoid over-smoothing of the details. The smaller the gradient magnitude of a region, the flatter the region is and more susceptible to noise. A larger value of K should be used to suppress the noise.

[0048] Fuse the sub-regions at the same position using the cross-scale fusion weights. The expression is:

[0049] In the formula, represents the luminance value of cross-scale fusion at position in the fusion region, represents the number of second feature maps, represents the cross-scale fusion weight at position in the sub-region of the p-th second feature map, Indicates the luminance value at the position in the p-th sub-region of the second feature map, where p represents the index of the second feature map. The luminance value at the position , and p represents the index of the second feature map.

[0050] Traverse each position to achieve cross-scale fusion of the second feature maps after adaptive filtering of three different scales, obtaining an enhanced terrain image, which can better protect local details and achieve cross-scale residual noise processing. It should be noted that filtering the sub-regions first and then fusing can retain more detailed information and is suitable for scenarios with high requirements for the enhanced image.

[0051] In one embodiment, first use the cross-scale fusion weights to fuse the sub-regions at the same position to obtain a preliminarily enhanced terrain image, and then use the anisotropic diffusion filtering algorithm to filter the enhanced terrain image. It should be noted that this filtering and denoising method has less computational complexity and is suitable for scenarios with relatively low requirements for the enhanced image.

[0052] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for enhancing terrain survey images for modular substation placement, characterized in that Including the steps: Obtain a terrain image, and perform convolution processing on the terrain image to obtain first feature maps of different scales; Adaptively adjust the size of the normalization window in the image local contrast normalization algorithm, and use the image local contrast normalization algorithm to perform normalization processing on the first feature maps of each scale to obtain second feature maps; Divide each second feature map into multiple sub-regions, and for multiple second feature maps, the sub-regions at the same position correspond to each other; Calculate the cross-scale fusion weight of each sub-region in the second feature map, and the cross-scale fusion weight is positively correlated with the information entropy of the sub-region; use the cross-scale fusion weight to fuse the sub-regions at the same position to obtain a fusion region, and multiple fusion regions constitute the enhanced terrain image.

2. A method for enhancing topographic survey images for modular substation placement according to claim 1, characterized in that, The method further includes: calculating the adjustment coefficient of the normalization window size, and the expression is: In the formula, represents the adjustment coefficient of the local contrast normalization window size of the pixel , represents the gradient magnitude of the pixel , represents the standard deviation of the brightness values of the neighboring pixels of the pixel , represents the maximum gradient magnitude of the pixels in the scale feature map, represents the maximum value of the standard deviation of the brightness values of the neighboring pixels of all the pixels in the scale feature map.

3. A method for enhancing topographic survey images for modular substation placement according to claim 2, characterized in that, The size of the normalization window in the image local contrast normalization algorithm is adaptively adjusted to: Wherein, represents a pixel point the adjusted normalized window represents the size of a preset window represents a pixel point the adjustment coefficient of the local contrast normalization window size of the image of, exp represents the exponential function with base e represents the floor symbol.

4. A topographic survey image enhancement method for modular substation placement according to claim 1, characterized in that The expression of the cross-scale fusion weight is: In the formula, represents the cross-scale fusion weight of the sub-region in the p-th second feature map, represents the information entropy of the brightness value of the sub-region in the p-th second feature map, where p represents the index of the second feature map, represents the number of second feature maps.

5. A method for enhancing topographic survey images for modular substation placement according to claim 1, characterized in that Before using the cross-scale fusion weight to fuse the sub-regions at the same position, it further includes: calculating the thermal conduction coefficient in the anisotropic diffusion filtering algorithm, and using the anisotropic diffusion filtering algorithm to filter the sub-regions in the second feature map.

6. A topographic survey image enhancement method for modular substation placement according to claim 5, characterized in that, The expression of the thermal conduction coefficient is: In the formula, represents the heat conduction coefficient of sub-region adaptive adjustment in the second feature map at scale . represents the preset heat conduction coefficient. represents the average value of the gradients of all pixel points in the sub-region of the second feature map at scale in the axis direction. represents the average value of the gradients of all pixel points in the sub-region of the second feature map at scale in the axis direction. represents the maximum value of the gradients in the sub-region.

7. A method for enhancing topographic survey images for modular substation installation according to claim 6, characterized in that, The expression of using the cross-scale fusion weight to fuse the sub-regions at the same position is: In the formula, represents the luminance value of cross-scale fusion at the position in the fusion region, represents the number of second feature maps, represents the cross-scale fusion weight at the position in the sub-region of the p-th second feature map, represents the luminance value at the position in the sub-region of the p-th second feature map, where p represents the index of the second feature map.

8. A topographic survey image enhancement method for modular substation placement according to claim 1, characterized in that, The method further includes using the anisotropic diffusion filtering algorithm to filter the enhanced terrain image.

9. A method for enhancing topographic survey images for modular substation placement according to claim 1, characterized in that, Before performing convolution processing on the terrain image, it further includes: performing geometric correction and radiometric correction on the terrain image.

10. A topographic survey image enhancement method for modular substation placement according to claim 1, characterized in that, Before using the image local contrast normalization algorithm to perform normalization processing on the first feature maps of each scale, it further includes: using bilinear interpolation upsampling for the first feature maps of different scales to make the size of the first feature maps the same as that of the terrain image.

Images