Defogging method for UAV aerial images based on fusion of atmospheric light value-map estimation

Abstract

The present invention discloses a method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation, obtaining a UAV aerial image containing fog I; converting the foggy image I into an HSV image, and obtaining an image depth map I by subtracting the V channel from the S channel. d ; Image depth map I d Perform minimum deviation processing to obtain the atmospheric light value A c ; Obtain dark channel map for foggy image, perform adaptive random walk clustering, and obtain atmospheric light map A RW ; The atmospheric light value A c And the atmospheric light map A RW The fusion process is used to estimate atmospheric light; transmittance is calculated based on the haze line prior and the obtained atmospheric light estimate, and then the transmittance is optimized through dark compensation. Based on the atmospheric scattering model, the dehazed image is calculated by fusing the atmospheric light estimate and the optimized transmittance. Based on the atmospheric scattering model, this paper proposes a method for fusing atmospheric light value maps to compensate for the shortcomings of a single atmospheric light value map. Furthermore, dark compensation is used to optimize transmittance, resulting in a more accurate atmospheric light map and transmittance.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method for defogging unmanned aerial vehicle (UAV) aerial images based on fusion of atmospheric light value-map estimation. Background Art

[0002] The frequent occurrence of smog severely impacts public health and poses significant challenges to the processing reliability of various visual systems. Foggy images often suffer from low visibility and low color saturation. These images deviate significantly from the real scene and fail to accurately reflect scene information, severely impacting subsequent image processing and information extraction. Serious errors in the visual system's recognition and judgment processes can cause immeasurable losses in practical applications. Drone aerial imagery, a crucial tool for obtaining traffic information, road condition surveys and early warnings, and urban computing, is severely impacted by smog. Therefore, image dehazing is particularly important for improving the reliability of visual systems.

[0003] The primary goal of image dehazing technology is to remove the impact of haze on image quality. It primarily utilizes image enhancement methods or physical models to achieve this dehazing effect, improving image visibility and addressing issues such as low visibility, blurred edges, and low contrast. As a key image preprocessing technique, image dehazing effectively improves the quality of hazy images and serves as the foundation for image processing techniques such as pattern recognition, image classification, and segmentation. Therefore, research on image dehazing algorithms is crucial for improving the stability of machine vision systems. Due to the numerous factors that contribute to image degradation during image acquisition, the complex and varied image content, and the significant differences in scene depth, restored images often exhibit image distortion, dimming, increased noise, and "halo" artifacts. Further improving the quality of restored images is crucial for subsequent processing. Summary of the Invention

[0004] The purpose of the present invention is to provide a defogging method for UAV aerial images based on the fusion of atmospheric light value-map estimation, so as to solve the problems of low brightness of restored images and obvious color distortion of the sky in most existing defogging algorithms. Based on the atmospheric scattering model, the present invention proposes a method of fusing atmospheric light value-map to make up for the shortcomings of a single atmospheric light value, and optimizes the transmittance through dark compensation to obtain a more accurate atmospheric light map and transmittance.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation includes the following steps:

[0007] Step 1: Obtain foggy image I taken by drone;

[0008] Step 2: Convert the foggy image I into an HSV image and obtain the image depth map I by subtracting the V channel from the S channel. d ;

[0009] Step 3: Depth Map I d Perform minimum deviation processing to obtain the atmospheric light value A c ;

[0010] Step 4: Obtain the dark channel map of the foggy image and perform adaptive random walk clustering to obtain the atmospheric light map A RW ;

[0011] Step 5: The atmospheric light value A obtained in step 3 c The atmospheric light map A obtained in step 4 RW Fusion as fused atmospheric light estimation;

[0012] Step 6: Calculate the transmittance based on the haze line prior and the atmospheric light estimate obtained in step 5, and then optimize the transmittance through dark compensation;

[0013] Step 7: According to the atmospheric scattering model, the dehazed image is obtained by fusing the atmospheric light estimation and optimizing the transmittance calculation.

[0014] Furthermore, the image depth map I is obtained in step 2 d The method is:

[0015] Convert the foggy image I into an HSV image and subtract the V channel from the S channel to obtain the difference. The formula is as follows:

[0016] I d (x,y)=I v (x,y)-I s (x,y) (1)

[0017] Among them, I d (x, y) is the value of the image depth map at (x, y), I v (x,y) and I s (x, y) are the V channel and S channel corresponding to the pixel point (x, y) in the HSV image.

[0018] Furthermore, the atmospheric light value A is obtained in step 3 c The method is:

[0019] The image depth map I d Perform minimum deviation processing to obtain the minimum deviation area. The formula is as follows:

[0020]

[0021] Among them, M A(i, j) is the central area where the atmospheric light value is located, and Ω(i, j) is the depth map I d The local area centered at (i, j) is δ[·], which is the variance operation. A The average value of the pixel (i, j) is taken as the atmospheric light value A c , the formula is as follows:

[0022]

[0023] Among them, A c is the atmospheric light value, Σ[M A (i,j)] is the region M A (i, j) pixel value is summed, N a is the number of pixels in the area.

[0024] Furthermore, the adaptive random walk clustering in step 4 is specifically as follows:

[0025] Step 4.1: Sum the dark channel image row by row, and select N coordinate points in the row with the largest summed peak as seed points;

[0026] Step 4.2: Perform random walk processing on the obtained N seed points to obtain N sub-regions;

[0027] Step 4.3: Take the average of the first 0.1% of each sub-region as the atmospheric light of the sub-region;

[0028] Step 4.4: Combine the atmospheric light from each sub-region and then filter it through guided filtering to obtain the atmospheric light map A. RW .

[0029] Furthermore, the specific method of taking the coordinate point as the seed point in step 4.1 is:

[0030] Get the dark channel image I corresponding to the foggy image dark , as shown below

[0031]

[0032] Among them, I c For any channel in the RGB channel, sum the dark channel rows and select the N peak rows with the largest peak fluctuations. The formula is as follows

[0033] S r =f p (s r (I dark )) (5)

[0034] Among them, S r is the row number corresponding to the peak value, f p () is the function of taking the peak value, sr () is the row sum function. Then, traverse the peak row and find the column number corresponding to the peak point in the peak row. The formula is as follows

[0035] S c =f p (I dark (S r )) (6)

[0036] Among them, S c is the peak pixel column coordinate, I dark (S r ) is the S in the dark channel image r The row corresponds to the pixel. Through the above method, the peak seed point of the image is obtained. S r is the horizontal coordinate of the seed point, S c is the vertical coordinate of the seed point.

[0037] Furthermore, the specific method for calculating the atmospheric light in each sub-region in step 4.3 is:

[0038] For each sub-area Q1~Q N The average value of the first 0.1% is taken as the atmospheric light value of the area. The formula is as follows:

[0039]

[0040] Among them, A qi is the i-th sub-region Q i The atmospheric light value, For this sub-region Q i Sum the first 0.1% pixel values, n i For this sub-region Q i The number of pixels in the top 0.1%;

[0041] The atmospheric light value combination of each sub-region is filtered and the atmospheric light map A is obtained. RW , the formula is as follows

[0042] A RW =guidefilter(A q1 ∪A q2 ∪…∪A qN ) (8)

[0043] Among them, A RW is the atmospheric light map, guidefilter() is the guide filter function, A q1 , A q2 , A qN is the atmospheric light value corresponding to each sub-area.

[0044] Furthermore, the fused atmospheric light estimation in step 5 is calculated as follows:

[0045] IgA=lgA c +lg(1-A RW ) (9)

[0046] Among them, A is the fusion atmospheric light estimation, A c is the atmospheric light value obtained in step 3, A RW This is the atmospheric light map for step 4.

[0047] Furthermore, the method for calculating the transmittance in step 6 is:

[0048] According to the haze line theory, the fused atmospheric light estimate is substituted into the haze line model to obtain the transmittance t(x);

[0049] The specific method of the haze line model is:

[0050] Step 6.1: Convert the foggy image I into spherical coordinates using the following formula:

[0051]

[0052] Among them, I(x) is the foggy image, A is the fusion atmospheric light estimation, I A (x) is the difference between the foggy pixel value and the atmospheric light, r(x) is the radius distance of the pixel point in the spherical coordinate system after removing the atmospheric light brightness; θ(x) and I in spherical coordinate system A (x) elevation and azimuth angles;

[0053] Step 6.2: Cluster r(x) into different clusters and find the maximum radiation r within the cluster max ;

[0054] r max =||I(x m )-A|| (11)

[0055] Among them, r max is the maximum radiance within the cluster, x m is the pixel point not affected by haze, I(x m ) is the x in the foggy image I m The corresponding pixel value;

[0056] Step 6.3: Calculate the transmittance using the following formula:

[0057] t(x)=r(x) / r max (12)

[0058] Among them, t(x) is the transmittance, r(x) is the radius distance of the pixel point in the spherical coordinate system after removing the atmospheric brightness, r max is the maximum radiance within the cluster.

[0059] Furthermore, the method for optimizing transmittance by dark compensation in step 6 is:

[0060] The dark compensation formula is defined as follows:

[0061] t b (x)≈1-w·(I dark (x) / A) (13)

[0062] Among them, t b (x) is dark compensation, w is the adjustment coefficient, I dark (x) is the pixel value corresponding to pixel x in the dark channel image, and A is the fused large area light;

[0063] The dark compensation optimization method is as follows:

[0064]

[0065] in: is the transmittance after dark compensation optimization, th is the compensation threshold, dt is the difference between dark compensation and transmittance, which is negative, and t(x) is the transmittance before dark compensation optimization.

[0066] Furthermore, the atmospheric scattering model in step 7 is:

[0067]

[0068] Where: J(x) is the restored clear image, I(x) is the foggy image, and A is the fused atmospheric light map. Transmittance after optimization for dark compensation.

[0069] Compared with the prior art, the present invention has the following beneficial technical effects:

[0070] For the original foggy images obtained in hazy weather, traditional defogging algorithms often use a single atmospheric light value method to defog. Most of the images after defogging have problems such as low brightness of the restored image and obvious color distortion of the sky. Based on the atmospheric scattering model, the present invention makes improvements to the atmospheric light estimation and transmittance estimation respectively, and proposes a method of fusing atmospheric light value-map. The atmospheric light value of the area with the minimum deviation is fused with the atmospheric light map estimation based on random walk, which solves the influence of single atmospheric light on the local image and improves the accuracy of atmospheric light estimation. The overall brightness and color of the defogging image are greatly improved. After obtaining the transmittance using the haze line principle, the transmittance is optimized by dark compensation. This method not only retains the original information structure of the image, but also further improves the accuracy of the defogging image and greatly reduces the occurrence of color distortion. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The drawings in the specification are used to provide further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0072] Figure 1 It is a schematic flow diagram of the present invention;

[0073] Figure 2 This is a comparison of the defogging effects of the present invention and other traditional defogging algorithms on images under hazy weather, where (a) is the original image under hazy weather, (b) the defogging image based on the dark channel, (c) the defogging image based on the improved dark channel, (d) the defogging image based on the FVID method, (e) the defogging image based on the haze line, (f) the dehazing image based on DehazeNet, and (g) the final defogging image of the present invention. DETAILED DESCRIPTION

[0074] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0075] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0076] See also Figure 1The present invention provides a defogging method for drone aerial images based on fusion of atmospheric light value-map estimation. The method of fusion of atmospheric light value-map and dark compensation optimization of transmittance improves the defogging accuracy and obtains better defogging effect. The specific idea is: first, obtain the depth of field image, and take the first 0.1% mean of the area with the smallest deviation as the atmospheric light value; second, design an adaptive random walk clustering method to estimate the atmospheric light map, cluster the image into N sub-regions through the adaptive random walk algorithm, and obtain the atmospheric light map by averaging the first 0.1% of the sub-regions, and refine it through guided filtering; then, fusion of atmospheric light value-map estimation obtains a new atmospheric light map; the haze line prior method is used to obtain the transmittance, and a dark compensation method is proposed to optimize it; finally, according to the atmospheric scattering model, the obtained fused atmospheric light map and transmittance are used to obtain a clear restored image. The specific steps are as follows:

[0077] Step 1: Obtain foggy image I taken by drone;

[0078] Step 2: Convert the foggy image I obtained in step 1 into an HSV image, and obtain the image depth map I by subtracting the V channel from the S channel. d , the formula is as follows:

[0079] I d (x,y)=I v (x,y)-I s (x,y) (1)

[0080] Where, I v (x,y) and I s (x, y) are the V channel and S channel corresponding to the pixel point (x, y) in the HSV image.

[0081] Step 3: Perform minimum deviation processing on the image depth map Id obtained in step 2 to obtain the minimum deviation area. The formula is as follows:

[0082]

[0083] Where M A (i, j) is the central area where the atmospheric light value is located, and Ω(i, j) is the depth map I d The local area centered at (i, j) is δ[·], which is the variance operation. A The average value of the pixel (i, j) is taken as the atmospheric light value A c , the formula is as follows:

[0084]

[0085] Where A c is the atmospheric light value, Σ[M A(i,j)] is the region M A (i, j) pixel value is summed, N a is the number of pixels in the area.

[0086] Step 4: Get the dark channel image I corresponding to the image dark , as shown below

[0087]

[0088] Where, I c For any channel in the RGB channel, sum the dark channel rows and select the N peak rows with the largest peak fluctuations. The formula is as follows

[0089] S r =f p (s r (I dark )) (5)

[0090] Where S r is the row number corresponding to the peak value, f p () is the function of taking the peak value, s r () is the row sum function. Then, traverse the peak row and find the column number corresponding to the peak point in the peak row. The formula is as follows

[0091] S c =f p (I dark (S r )) (6)

[0092] Where S c is the peak pixel column coordinate, I dark (S r ) is the S in the dark channel image r The row corresponds to the pixel. Through the above method, the peak seed point of the image is obtained. S r is the horizontal coordinate of the seed point, S c is the ordinate of the seed point. The N seed points are randomly walked to obtain N sub-regions. The first 0.1% of each sub-region is taken as the average atmospheric light of the sub-region. The formula is as follows

[0093]

[0094] Where A qi is the i-th sub-region Q i The atmospheric light value, For this sub-region Q i Sum the first 0.1% pixel values, n i For this sub-region Q i The number of pixels in the top 0.1%.

[0095] The atmospheric light value combination of each sub-region is filtered and the atmospheric light map A is obtained. RW , the formula is as follows

[0096] A RW =guidefilter(A q1 ∪A q2 ∪…∪A qN ) (8)

[0097] Where A RW is the atmospheric light map, guidefilter() is the guide filter function, A q1 , A q2 , A qN is the atmospheric light value corresponding to each sub-area.

[0098] Step 5: The atmospheric light value A obtained in step 3 c The atmospheric light map A obtained in step 4 RW Fusion as fused atmospheric light estimation

[0099] IgA=lgA c +lg(1-A RW ) (9)

[0100] Where A is the fusion atmospheric light estimation, A c is the atmospheric light value obtained in step 3, A RW This is the atmospheric light map for step 4.

[0101] Step 6: Convert the foggy image I into spherical coordinates according to the haze line prior. The formula is as follows:

[0102]

[0103] Where I(x) is the foggy image, A is the fusion atmospheric light estimation, and I A (x) is the difference between the foggy pixel value and the atmospheric light, r(x) is the radius distance of the pixel point in the spherical coordinate system after removing the atmospheric light brightness; θ(x) and I in spherical coordinate system A (x) elevation and azimuth angles.

[0104] Cluster r(x) into different clusters and find the maximum radiation r within the cluster max .

[0105] r max =||I(x m )-A|| (11)

[0106] Where r max is the maximum radiance within the cluster, x m is the pixel point not affected by haze, I(xm ) is the x in image I m The corresponding pixel value;

[0107] To calculate transmittance, the formula is as follows:

[0108] t(x)=r(x) / r max (12)

[0109] Where t(x) is the transmittance, r(x) is the radius of the pixel point in the spherical coordinate system after removing the atmospheric brightness, and r max is the maximum radiance within the cluster.

[0110] The obtained transmittance is then optimized for dark compensation, and the dark compensation formula is defined as follows:

[0111] t b (x)≈1-w·(I dark (x) / A) (13)

[0112] Where, t b (x) is dark compensation, w is the adjustment coefficient, I dark (x) is the pixel value corresponding to pixel x in the dark channel image, and A is the fused large area light. The dark compensation optimization method is as follows:

[0113]

[0114] Where: is the transmittance after dark compensation optimization, th is the compensation threshold, dt is the difference between dark compensation and transmittance, which is negative, and t(x) is the transmittance before dark compensation optimization.

[0115] Step 7: Substitute the atmospheric scattering model to obtain the defogging image:

[0116]

[0117] Where: J(x) is the restored clear image, I(x) is the foggy image, A is the fused atmospheric light map, Transmittance after optimization for dark compensation.

[0118] from Figure 2 It can be seen that Figure 2(a) is the original image under haze weather, which is dehazed using the traditional dehazing algorithm and the random walk algorithm proposed in this paper respectively; b) The dehazed image obtained based on the dark channel method restores the information in the image well, and the details of the buildings on the other side of the river are relatively clear, but the overall image is darker, and the processing effect in the dense fog area is not obvious; (c) The dehazed image based on the improved dark channel has a certain improvement in brightness and contrast, and the sky area becomes darker; (d) The dehazed image based on the FVID method has a certain improvement in brightness, but the processing of the details of the distant buildings is worse than that of DCP; (e) The dehazed image based on the haze line process the architectural details well, but there is obvious distortion in some areas, especially the shadow parts, such as the green area on the other side of the river, the shadow part under the bridge, and the cargo ship in the river. (f) The dehazed image based on DehazeNet has improved brightness and contrast, but the overall effect is not outstanding, and the processing of distant objects is not obvious. (g) The final dehazed image of the present invention is significantly better than other algorithms. The brightness and contrast are significantly improved, and the processing of architectural details is also better. In particular, the buildings on the other side of the river are clearly visible, and the bottom of the bridge and the cargo ship are also more obvious.

[0119] Table 1 Objective evaluation indicators of dehazed images

[0120]

[0121] Table 1 compares the dehazed image performance using information entropy, average gradient, fuzziness coefficient, and contrast. As shown in Table 1, the proposed algorithm, due to the fusion of the atmospheric light value map, achieves superior atmospheric light estimation quality compared to other algorithms. The processed image is more colorful, with the highest contrast, average gradient, and fuzziness coefficient data. Information entropy is also improved, and the haze concentration in the restored image is significantly reduced, resulting in superior dehazing performance. This demonstrates that the proposed method is superior to other common dehazing algorithms.

[0122] Finally, it should be noted that the above embodiments are merely preferred embodiments of the present invention and are intended to illustrate the technical solutions of the present invention, rather than limiting them, and certainly not limiting the patent scope of the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention. In other words, any changes or refinements made to the main design concept and spirit of the present invention that have no substantive significance, provided that the technical problems they solve are still consistent with those of the present invention, should be included in the protection scope of the present invention. In addition, the direct or indirect application of the technical solutions of the present invention to other related technical fields should also be included in the patent protection scope of the present invention.

Claims

1. A method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation, characterized in that: The following steps are involved: Step 1: Obtain foggy image I taken by drone; Step 2: Convert the foggy image I into an HSV image and obtain the image depth map I by subtracting the V channel from the S channel. d ; Step 3: Depth Map I d Perform minimum deviation processing to obtain the atmospheric light value A c ; Step 4: Obtain the dark channel map of the foggy image and perform adaptive random walk clustering to obtain the atmospheric light map A RW ; Step 5: The atmospheric light value A obtained in step 3 c The atmospheric light map A obtained in step 4 RW Fusion as fused atmospheric light estimation; Step 6: Calculate the transmittance based on the haze line prior and the atmospheric light estimate obtained in step 5, and then optimize the transmittance through dark compensation; The method for calculating transmittance is: According to the haze line theory, the fused atmospheric light estimate is substituted into the haze line model to obtain the transmittance t(x); The specific method of the haze line model is: Step 6.1: Convert the foggy image I into spherical coordinates using the following formula: Among them, I(x) is the foggy image, A is the fusion atmospheric light estimation, I A (x) is the difference between the foggy pixel value and the atmospheric light, r(x) is the radius distance of the pixel point in the spherical coordinate system after removing the atmospheric light brightness; θ(x) and φ(x) are the I A (x) elevation and azimuth angles; Step 6.2: Cluster r(x) into different clusters and find the maximum radiation r within the cluster max ; r max =||I(x m )-A|| (11) Among them, r max is the maximum radiance within the cluster, x m is the pixel point not affected by haze, I(x m ) is the x in the foggy image I m The corresponding pixel value; Step 6.3: Calculate the transmittance using the following formula: t(x)=r(x) / r max (12) Among them, t(x) is the transmittance, r(x) is the radius distance of the pixel point in the spherical coordinate system after removing the atmospheric brightness, r max is the maximum radiance within the cluster; Step 7: According to the atmospheric scattering model, the dehazed image is obtained by fusing the atmospheric light estimation and optimizing the transmittance calculation.

2. The method for defogging unmanned aerial images based on fusion of atmospheric light value-map estimation according to claim 1, characterized in that: The image depth map I is obtained in step 2 d The method is: Convert the foggy image I into an HSV image and subtract the V channel from the S channel to obtain the difference. The formula is as follows: I d (x,y)=I v (x,y)-I s (x,y) (1) Among them, I d (x, y) is the value of the image depth map at (x, y), I v (x,y) and I s (x, y) are the V channel and S channel corresponding to the pixel point (x, y) in the HSV image.

3. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 1, characterized in that: Obtain atmospheric light value A in step 3 c The method is: The image depth map I d Perform minimum deviation processing to obtain the minimum deviation area. The formula is as follows: Among them, M A (i, j) is the central area where the atmospheric light value is located, and Ω(i, j) is the depth map I d The local area centered at (i, j) is δ[·], which is the variance operation. A The average value of the pixel (i, j) is taken as the atmospheric light value A c , the formula is as follows: Among them, A c is the atmospheric light value, Σ[M A (i,j)] is the region M A (i, j) pixel value is summed, N a is the number of pixels in the area.

4. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 1, characterized in that: The adaptive random walk clustering in step 4 is specifically as follows: Step 4.1: Sum the dark channel image and select N coordinate points in the row with the largest summed peak as seed points; Step 4.2: Perform random walk processing on the obtained N seed points to obtain N sub-regions; Step 4.3: Take the average of the first 0.1% of each sub-region as the atmospheric light of the sub-region; Step 4.4: Combine the atmospheric light from each sub-region and then filter it through guided filtering to obtain the atmospheric light map A. RW .

5. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 4, characterized in that: The specific method of taking the coordinate point as the seed point in step 4.1 is: Get the dark channel image I corresponding to the foggy image dark , as shown below Among them, I c For any channel in the RGB channel, sum the dark channel rows and select the N peak rows with the largest peak fluctuations. The formula is as follows S r =f p (s r (I dark )) (5) Among them, S r is the row number corresponding to the peak value, f p () is the function of taking the peak value, s r () is the row sum function. Then, traverse the peak row and find the column number corresponding to the peak point in the peak row. The formula is as follows S c =f p (I dark (S r )) (6) Among them, S c is the peak pixel column coordinate, I dark (S r ) is the S in the dark channel image r The row corresponds to the pixel. Through the above method, the peak seed point of the image is obtained. S r is the horizontal coordinate of the seed point, S c is the vertical coordinate of the seed point.

6. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 4, characterized in that: The specific method for calculating the atmospheric light in each sub-region in step 4.3 is: For each sub-area Q1~Q N The average value of the first 0.1% is taken as the atmospheric light value of the area. The formula is as follows: Among them, A qi is the i-th sub-region Q i The atmospheric light value, For this sub-region Q i Sum the first 0.1% pixel values, n i For this sub-region Q i The number of pixels in the top 0.1%; The atmospheric light value combination of each sub-region is filtered and the atmospheric light map A is obtained. RW , the formula is as follows A RW =guidefilter(A q1 ∪A q2 ∪…∪A qN ) (8) Among them, A RW is the atmospheric light map, guidefilter() is the guide filter function, A q1 , A q2 , A qN is the atmospheric light value corresponding to each sub-area.

7. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 1, characterized in that: The fused atmospheric light estimation in step 5 is calculated as follows: lgA=lgA c +lg(1-A RW ) (9) Among them, A is the fusion atmospheric light estimation, A c is the atmospheric light value obtained in step 3, A RW This is the atmospheric light map for step 4.

8. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 1, characterized in that: The method for optimizing transmittance by dark compensation in step 6 is: The dark compensation formula is defined as follows: t b (x)≈1-w·(I dark (x) / A)(13) Among them, t b (x) is dark compensation, w is the adjustment coefficient, I dark (x) is the pixel value corresponding to pixel x in the dark channel image, and A is the fused large area light; The dark compensation optimization method is as follows: in: is the transmittance after dark compensation optimization, th is the compensation threshold, dt is the difference between dark compensation and transmittance, which is negative, and t(x) is the transmittance before dark compensation optimization.

9. The method for defogging UAV aerial images based on fusion of atmospheric light value-map estimation according to claim 8, characterized in that: The atmospheric scattering model in step 7 is: Where: J(x) is the restored clear image, I(x) is the foggy image, and A is the fused atmospheric light map. Transmittance after optimization for dark compensation.

Images