Insulator hydrophobicity image edge detection algorithm under CPS system considering deconvolution deblurring algorithm

Abstract

The invention belongs to the technical field of digital image processing, and particularly relates to an insulator hydrophobicity image edge detection algorithm under a CPS system considering a deconvolution deblurring algorithm, and the guarantee of insulator performance and the safe development of power grid construction cannot be separated. Hydrophobicity serves as an important parameter for detecting the performance of an insulator and is particularly important for an image edge detection algorithm of the insulator. According to the method, a deconvolution deblurring algorithm is combined, and a traditional edge detection method is improved. In order to solve the problem that image restoration is not ideal, the robustness of hydrophobic image blurring core estimation is enhanced by adopting an adaptive weight. In addition, the convergence of the algorithm is improved, the operation time is controlled, and the practicability and feasibility of the algorithm are further realized. The average operation time is shortened by about two thirds. And the edge detection algorithm under the deconvolution deblurring algorithm ignores the step, so that the convergence of the algorithm in the edge detection of the hydrophobic image is greatly improved, and the hydrophobic image can be quickly and accurately output.

Description

Technical Field

[0001] The invention belongs to the technical field of digital image processing, and in particular relates to an insulator hydrophobicity image edge detection algorithm in a CPS system taking into account a deconvolution deblurring algorithm. Background Art

[0002] With the mature development of information construction and the expansion of the scale of power grid, the power system is constantly improving. As one of the important equipment for system line insulation, the quality and safety performance of insulators have received more and more attention. However, with the passage of operating time, insulators will age to varying degrees under the influence of external environmental factors, which not only seriously affects their application value, but also easily causes accidents. Therefore, their hydrophobicity must be tested regularly. However, due to the problems of poor contrast, poor shadow, and uneven illumination in the traditional hydrophobic image edge detection algorithm, it is difficult to obtain good results in image processing. Therefore, it is urgent to establish an effective image edge detection algorithm to protect the safety of power grid lines. The deconvolution deblurring algorithm is one of the image restoration technology algorithms. It can not only obtain the maximum amount of valuable information in the image, but also has good image processing effects. It has been widely used in science and engineering. Applying it to insulator detection can improve the accuracy of hydrophobicity measurement, which has practical value for improving the performance of insulator hydrophobic image edge detection algorithm.

[0003] Insulators are used for general inspection and control of overhead distribution lines and locomotive top insulation, and are an important means to ensure the normal operation of the entire distribution line system and locomotive power system. Its excellent electrical performance is hydrophobicity, but the hydrophobicity of insulators is not constant. As the working time of insulators increases, the hydrophobicity of their surface will become worse and worse.

[0004] Generally speaking, there are five reasons for the decrease in the hydrophobicity of insulators: 1. Humidity: A sharp drop in surface resistance will lead to a sharp increase in leakage current; 2. Pollutant composition and accumulation level: The hydrophobicity decreases due to dirt accumulation; 3. Low temperature: Low temperature reduces the Brownian motion of LMW molecules; 4. Quality: Quality affects hydrophobicity; 5. Aging: Aging will deteriorate the composition of the insulator; It can be seen that with the increase in operating time, the insulator will show different degrees of aging, and its hydrophobicity will also decrease to different degrees. Under the current reality of a large area and a large population, the number of insulators used in power construction is also very large, and the resulting flashover accidents will still occur. Therefore, it is necessary to timely identify the hydrophobicity of insulators through image detection to avoid accidents and maintain the normal operation of the power system.

[0005] Due to the influence of different external factors, hydrophobic images often have certain defects, which makes the extraction effect of water droplets in the image not obvious. In view of this characteristic, the edge detection algorithm needs to perform certain preprocessing to obtain accurate images for subsequent classification. Generally speaking, the image processing in the hydrophobic image edge detection method is divided into three steps, namely, enhancing contrast, eliminating the reflection of water droplets, and segmenting the part with attached water droplets.

[0006] Enhance contrast: In the hydrophobic image of the insulator, the contrast between the water droplet and the insulator background is very low, which makes it very difficult to extract the water droplet. When performing image processing, the algorithm usually first converts each color image into a grayscale image, that is, extracts the brightness component for subsequent work, such as Figure 4 The process of converting a real color image to a grayscale image is to weight the R, G, and B components of each pixel in the real color image proportionally and add them together to obtain a two-dimensional image. This conversion method can meet the needs of most image processing, but it cannot meet the requirements of improving the low contrast of hydrophobic images.

[0007] Through relevant statistics, it is found that in the hydrophobic image of the insulator, whether it is a water droplet or a water film, except for the reflection point or reflection area, the R component value is almost the same as the surrounding background. However, the value of the G group is 30 lower than the background pixel value, and the value of the B group is 40 lower than the background pixel value. The R, G, and B components of the reflection point are almost close to 240. Whether it is a water droplet or a background pixel, the value of the R component is the largest, and the B component is always the smallest. However, when the real color image is converted to a grayscale image, the B component with the largest difference is assigned the smallest value, so the final grayscale image has a small contrast, such as Figure 5 As shown; Figure 5 The positions marked with numbers in the figure represent the areas with five different pixel features in the hydrophobic image, 1 represents the water film area, 2 represents the reflection area, 3 represents the water drop area, 4 represents the background area, and 5 represents the reflection point area. Table 2 lists the Figure 2 10 pixels in .

[0008] Table 1: Contrast between water droplet pixels and background pixels

[0009]

[0010] Finally, eliminate the reflection of water droplets In the edge detection of hydrophobic images of insulators, the optical properties of water droplets and films are different. When the waterproof level of the insulator is low, the better the waterproof effect, the more water droplets, and the larger the contact angle between the water droplets and the insulator. At this time, the water droplet is like a convex lens, and under the irradiation of strong light, there will be obvious reflection points. This bright spot or bright area is brighter than the surrounding water droplets, and its pixel value component is much larger than other parts of the water droplet, which affects the subsequent extraction of water droplets. When the water repellent level of the insulator is high, the water mark gradually becomes the entire water film. At this time, the contact angle between the water film and the insulator is very small, so there are few or no reflection points. Under normal circumstances, the reflection points or reflection areas of water droplets are regarded as uneven lighting. Image edge detection algorithms usually use histogram equalization to eliminate the influence of water droplet reflection. When the brightness difference between the reflection point or reflection area and the surrounding pixel values ​​is small, it can play a certain role, but when the difference is large, it basically does not work.

[0011] Divide the attached water droplets; the adhesion of water droplets refers to the slight contact between the edges of two water droplets, which appears as two white areas connected by one or two pixels in the binary image. In the process of judging the hydrophobicity level of the insulator, the image edge detection method generally judges its hydrophobicity based on the characteristic value of the largest water droplet or water film in the image. If the sum of the areas of the two adhered water droplets in the image is larger than the area of ​​the single largest water droplet, the two adhered water droplets are regarded as one water droplet and their characteristic parameters are extracted. However, the real largest water droplet is ignored, and errors will occur at this time. For binary images, the narrow connection between two primitives is usually eliminated through morphological opening operations, such as Figure 6 As shown in the figure, for water drop images with large water drop areas and fewer attached parts, the algorithm can meet the requirements. However, for water drop images with small water drop areas and more attached parts, the algorithm cannot meet the requirements, the shape of the water drop will not be accurate enough, and even some water droplets may not be extracted.

[0012] In summary, the present invention proposes an edge detection algorithm for hydrophobic images of insulators under a CPS system taking into account a deconvolution deblurring algorithm. Aiming at the defects of the traditional edge detection algorithm, the characteristics of the hydrophobic image are re-analyzed, and the deconvolution deblurring algorithm is used for calculation. Through blur kernel estimation and sharp edge recovery, we have effectively improved this problem and also solved the problem of clear hydrophobic images under CPS. The practicality and effectiveness of the deconvolution deblurring algorithm were verified through experiments. Experimental data show that compared with the traditional edge detection algorithm, the comprehensive PSNR value of the insulator hydrophobic image edge detection algorithm based on deconvolution and deblurring algorithms is 27.02, and the improvement range is about 0.3dB to 1.9dB. In terms of running time, the average running time of the algorithm is reduced by about two-thirds. Summary of the invention

[0013] The purpose of the present invention is to provide an insulator hydrophobicity image edge detection algorithm under a CPS system that takes into account a deconvolution deblurring algorithm. The guarantee of insulator performance is inseparable from the safe development of power grid construction. As an important parameter for detecting insulator performance, hydrophobicity is particularly important for its image edge detection algorithm. The present invention improves the traditional edge detection method by combining it with a deconvolution deblurring algorithm. In view of the problems of poor edge detection effect and unsatisfactory image restoration in the original algorithm, adaptive weights are used to enhance the robustness of the hydrophobic image blur core estimation. In addition, the convergence of the algorithm is improved, the running time is controlled, and the practicality and feasibility of the algorithm are further realized.

[0014] The technical solution adopted by the present invention is as follows:

[0015] The edge detection algorithm of the insulator hydrophobicity image under the CPS system taking into account the deconvolution deblurring algorithm includes the following steps:

[0016] Step 1: Modeling the formation of the hydrophobicity image of the insulator: The modeling process is expressed as:

[0017] B=I*k+ε (1)

[0018] Where: k is the blur kernel, I is the latent graph, * is the franchise operation, ε is the random noise in the graph, and B is the hydrophobic image acquired by the imaging system; that is, the hydrophobic image can be regarded as a blur kernel, convolving a clear latent graph, and then adding random noise to obtain it;

[0019] Step 2: Most deconvolution and deblurring algorithms for hydrophobic images mainly utilize the sharp edges of hydrophobic images to help estimate the blur kernel; blur kernel estimation is performed on high-frequency images to restore the edges of the insulator hydrophobic image B. The fuzzy kernel estimation modeling formula is expressed as:

[0020]

[0021] Where: y represents the hydrophobic image The horizontal and vertical gradients of x represent the clear image. The horizontal and vertical gradients of x; k represents the blur kernel, ||·||0 represents the l0 norm, and its function is to calculate the number of non-zero values ​​in x;

[0022] Formula 2 contains three terms; the first term is the likelihood term, that is, the convolution of the restored clear hydrophobic image x and the blur kernel k should be similar to the observed hydrophobic map y; the second term is the l2-norm regularization term applied to the blur kernel, which is used to stabilize the estimate of the blur kernel; the third term is the l0-norm regularization term;

[0023] The solution of blur kernel is carried out in a multi-scale solution framework; the solution steps at each scale can be divided into blur kernel estimation and sharp edge recovery, which can be expressed as:

[0024]

[0025] In the blur kernel estimation stage, the blur kernel can be obtained by formula (3); since the blur kernel estimation is performed on the high-frequency part of the image, x represents the horizontal and vertical gradient maps of the potential sharp image, and y represents the horizontal and vertical gradient maps of the blurred image; the formula can be expressed as:

[0026]

[0027] Expand the formula to:

[0028]

[0029] In order to simplify the calculation, the fast Fourier transform (FFT) is used to transform the convolution operation in the formula into the multiplication of corresponding points in the frequency domain, and the derivative of k is calculated to solve formula (5) and obtain the minimum value of k, as shown in formula (6):

[0030]

[0031] In the step of restoring the sharp edges of the hydrophobic image, the terms related to k are removed, and formula (7) is obtained:

[0032]

[0033] Formula (7) is a non-convex problem with non-differentiable points. It cannot be solved by traditional optimization methods based on gradient descent. If a brute force search method is used, the solution process is very time-consuming. The present invention adopts an alternating direction method to solve the problem in formula (7).

[0034] Step 3: Introduce the auxiliary variable w and transform formula (7) into an equivalent problem in formula (8):

[0035]

[0036] Then we get the Lagrangian function, as shown in formula (9), where J1 and J2 represent Lagrangian multipliers and u is a penalty parameter:

[0037]

[0038] Therefore, the corresponding alternative iteration scheme is:

[0039]

[0040] The minimum value of can be converted into the problem shown in Formula 11:

[0041]

[0042] The solution to this problem is as follows:

[0043]

[0044] Step 4: In the deconvolution image blurring algorithm, only large-scale hydrophobic image gradient information is beneficial to blur kernel estimation; in the deconvolution image deblurring algorithm, the adaptive weight method is used to enhance the robustness of the blur kernel estimation of the hydrophobic image; the large-scale gradient information that has a positive contribution to the blur kernel estimation is screened, and the small image gradient information that has a negative contribution to the blur kernel estimation is eliminated; the improved model is shown in formula (13):

[0045]

[0046] Where k = exp(-|γ| 0.8 ),γ is defined as:

[0047]

[0048] Where B is the hydrophobic image of the insulator, N n (p) is a h*h window centered at each pixel; for small-scale edge information, due to the gradient image During the calculation process, these information will cancel each other out, causing γ to be too small, thus eliminating the small-scale edges that are not conducive to blur kernel estimation.

[0049] The technical effects achieved by the present invention are:

[0050] In the present invention, the guarantee of insulator performance is inseparable from the safe development of power grid construction; hydrophobicity, as an important parameter for detecting insulator performance, is particularly important for its image edge detection algorithm. The present invention combines the deconvolution deblurring algorithm to improve the traditional edge detection method. In view of the problems of poor edge detection effect and unsatisfactory image restoration in the original algorithm, the robustness of the hydrophobic image blur core estimation is enhanced by using adaptive weights. In addition, the convergence of the algorithm is improved, the running time is controlled, and the practicality and feasibility of the algorithm are further realized.

[0051] In this invention, the characteristics of the hydrophobic image are re-analyzed to address the defects of the traditional edge detection algorithm, and the deconvolution deblurring algorithm is used for calculation; through blur kernel estimation and sharp edge recovery, the invention effectively improves this problem and solves the problem of clear hydrophobic images under CPS; through experiments, the practicality and effectiveness of the deconvolution deblurring algorithm are verified. Experimental data show that compared with the traditional edge detection algorithm, the comprehensive PSNR value of the insulator hydrophobic image edge detection algorithm based on the deconvolution and deblurring algorithm is 27.02, and the improvement range is about 0.3dB to 1.9dB; in terms of running time, the average running time of the algorithm is reduced by about two-thirds. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a comparison diagram of the results of using the edge detection algorithm of the present invention and the traditional edge detection algorithm;

[0053] Figure 2 is the objective evaluation result of the hydrophobic image in the present invention;

[0054] Figure 3 It is the result of the operation time of the algorithm in the present invention;

[0055] Figure 4 It is the background technology of the present invention to convert a color image into a grayscale image;

[0056] Figure 5 It is a region block with different pixel characteristics in the background technology of the present invention;

[0057] Figure 6 It is an example diagram of the opening operation process in the background technology of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present invention, and does not strictly limit the scope of protection of the specific claims of the present invention.

[0059] like Figure 1 As shown, the edge detection algorithm of the insulator hydrophobicity image under the CPS system taking into account the deconvolution deblurring algorithm includes the following steps:

[0060] Step 1: Modeling the formation of the hydrophobicity image of the insulator: The modeling process is expressed as:

[0061] B=I*k+ε (1)

[0062] Where: k is the blur kernel, I is the latent graph, * is the franchise operation, ε is the random noise in the graph, and B is the hydrophobic image acquired by the imaging system; that is, the hydrophobic image can be regarded as a blur kernel, convolving a clear latent graph, and then adding random noise to obtain it;

[0063] Step 2: Most deconvolution and deblurring algorithms for hydrophobic images mainly utilize the sharp edges of hydrophobic images to help estimate the blur kernel; blur kernel estimation is performed on high-frequency images to restore the edges of the insulator hydrophobic image B. The fuzzy kernel estimation modeling formula is expressed as:

[0064]

[0065] Where: y represents the hydrophobic image The horizontal and vertical gradients of x represent the clear image. The horizontal and vertical gradients of x; k represents the blur kernel, ||·||0 represents the l0 norm, and its function is to calculate the number of non-zero values ​​in x;

[0066] Formula 2 contains three terms; the first term is the likelihood term, that is, the convolution of the restored clear hydrophobic image x and the blur kernel k should be similar to the observed hydrophobic map y; the second term is the l2-norm regularization term applied to the blur kernel, which is used to stabilize the estimate of the blur kernel; the third term is the l0-norm regularization term;

[0067] The solution of blur kernel is carried out in a multi-scale solution framework; the solution steps at each scale can be divided into blur kernel estimation and sharp edge recovery, which can be expressed as:

[0068]

[0069] In the blur kernel estimation stage, the blur kernel can be obtained by formula (3); since the blur kernel estimation is performed on the high-frequency part of the image, x represents the horizontal and vertical gradient maps of the potential sharp image, and y represents the horizontal and vertical gradient maps of the blurred image; the formula can be expressed as:

[0070]

[0071] Expand the formula to:

[0072]

[0073] In order to simplify the calculation, the fast Fourier transform (FFT) is used to transform the convolution operation in the formula into the multiplication of corresponding points in the frequency domain, and the derivative of k is calculated to solve formula (5) and obtain the minimum value of k, as shown in formula (6):

[0074]

[0075] In the step of restoring the sharp edges of the hydrophobic image, the terms related to k are removed, and formula (7) is obtained:

[0076]

[0077] Formula (7) is a non-convex problem with non-differentiable points. It cannot be solved by traditional optimization methods based on gradient descent. If a brute force search method is used, the solution process is very time-consuming. The present invention adopts an alternating direction method to solve the problem in formula (7).

[0078] Step 3: Introduce the auxiliary variable w and transform formula (7) into an equivalent problem in formula (8):

[0079]

[0080] Then we get the Lagrangian function, as shown in formula (9), where J1 and J2 represent Lagrangian multipliers and u is a penalty parameter:

[0081]

[0082] Therefore, the corresponding alternative iteration scheme is:

[0083]

[0084] The minimum value of can be converted into the problem shown in Formula 11:

[0085]

[0086] The solution to this problem is as follows:

[0087]

[0088] Step 4: In the deconvolution image blurring algorithm, only large-scale hydrophobic image gradient information is beneficial to blur kernel estimation; in the deconvolution image deblurring algorithm, the adaptive weight method is used to enhance the robustness of the blur kernel estimation of the hydrophobic image; the large-scale gradient information that has a positive contribution to the blur kernel estimation is screened, and the small image gradient information that has a negative contribution to the blur kernel estimation is eliminated; the improved model is shown in formula (13):

[0089]

[0090] Where k = exp(-|γ| 0.8 ),γ is defined as:

[0091]

[0092] Where B is the hydrophobic image of the insulator, N n (p) is a h*h window centered at each pixel; for small-scale edge information, due to the gradient image During the calculation process, these information will cancel each other out, causing γ to be too small, thus eliminating the small-scale edges that are not conducive to blur kernel estimation.

[0093] In the present invention, in order to test the effectiveness of the insulator hydrophobic image edge detection method considering the deconvolution deblurring algorithm, the present invention tests some standard insulator hydrophobic standard images, and the image information is shown in Table 2:

[0094] Table 2: Basic information of the test images

[0095]

[0096] In the present invention, image detection is performed; in the test of the present invention, in order to verify the implementation of the algorithm under degraded conditions, the present invention adopts a degradation model of the insulator hydrophobic image and adds Gaussian white noise with a standard deviation of 0.5. First, the insulator hydrophobic degraded image is processed using a traditional image edge detection algorithm and an i image edge detection method that considers deconvolution and deblurring methods, and the detection results are compared. The parameters of each algorithm are set to the optimal value through experimental methods. The test results are shown in Figure 2. Figure 1 As shown: Figure 1 A describes the results of the edge detection algorithm in the present invention; Figure 1 B describes the results of the traditional edge detection algorithm; from the edge detection algorithm of the present invention and Figure 1 It can be clearly seen from the results of the traditional method in the figure that the ripple effect of the hydrophobic image of the insulator in the traditional algorithm is more serious and the edge detection effect is poor. The reason for this phenomenon is that the model established by the traditional edge detection algorithm has a large deviation from the original hydrophobic image, which causes the algorithm to deviate from the actual optimal in the iterative process. Although the algorithm adjusts this deviation by reconstructing the model, since the new model is based on intermediate estimates, each estimate is obtained by the same method, which makes it easy for some defects in the image to solidify continuously, resulting in an increase in ripples and difficulty in edge recovery close to the original hydrophobic image.

[0097] The edge detection algorithm proposed in the present invention taking into account the deconvolution deblurring algorithm has solved this problem well. By comparison, it is found that the difference between the result of the algorithm of the present invention and the original hydrophobic image is generally small. The result of the algorithm proposed in the present invention maintains a better restoration effect of the texture part, has a clearer edge, and also overcomes the problems of image reflection and low contrast in traditional edge detection algorithms.

[0098] In the present invention, in order to objectively evaluate the restoration performance, the present invention uses peak signal-to-noise ratio (PSNR) and relative residual error (RRE) as evaluation indicators. Although these two evaluation criteria have different forms, they essentially use the difference between the original hydrophobic map and the estimated hydrophobic map as the basis for evaluation. The algorithm performance evaluation method used in this experiment is mainly based on these two objective quantitative evaluation criteria, supplemented by subjective visual evaluation.

[0099] like Figure 2 As shown, Figure 2 A shows the PSNR results; Figure 2 B is the result of RRE; Figure 2 It can be clearly seen from the objective evaluation indicators in that for various test images, the improved edge detection method of the present invention has a better PSNR ratio than the traditional method. Under the method of the present invention, the comprehensive PSNR ratio of the six hydrophobic images is 27.02, while the comprehensive PSNR ratio of the conventional edge detection method is 26.33, and the improvement range is about 0.3dB to 1.9dB. Combined with subjective visual evaluation, it is shown that the method is effective. In the RRE results, the RRE value of the present invention is relatively small, with a comprehensive value of only 6.55, while the average RRE value of the traditional edge detection algorithm reaches 6.98. From this result, it can be clearly seen that the algorithm has a better fitting result, making the obtained image smoother and having a better noise suppression effect. However, the traditional insulator hydrophobic edge detection algorithm is damaged to a certain extent during image processing. The reason is that the images used in the experiment are all natural hydrophobic images. One of the characteristics of natural hydrophobic images is that their pixel grayscale and distribution lack regularity. That is to say, the correlation between them is not strong. The improved edge detection method of the present invention under the deconvolution deblurring algorithm has a strong directionality, and its advantages are well reflected.

[0100] In the present invention, in terms of operation time, the running time required for the two algorithms to achieve convergence on different hydrophobic images under the degradation model is compared. The results are as follows: Figure 3 As shown: Figure 3 A is the running time of the edge detection method of the present invention; Figure 3 B shows the operation time of the traditional edge detection method; Figure 3From the comparison results of the operation time of the algorithm in different edge sub-hydrophobic images, it can be seen that the method has ideal performance in the negative aspect of convergence speed. For 6 different types of hydrophobic images, considering the edge detection algorithm under the deconvolution deblurring algorithm, the comprehensive operation time is 244.53s. Compared with the comprehensive running time of 589.77s of the traditional edge detection algorithm, the average running time of the present invention is shortened by about two-thirds. Generally speaking, when edge detection is performed on the hydrophobic image of the insulator, due to the inherent characteristics of the graphics, the algorithm often needs to repeatedly train the model. However, the edge detection algorithm under the deconvolution deblurring algorithm ignores this step, which greatly improves the convergence of the algorithm in the edge detection of the hydrophobic image, enabling it to output quickly and accurately.

[0101] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. An insulator hydrophobicity image edge detection algorithm in a CPS system taking into account a deconvolution deblurring algorithm, characterized by: The following steps are involved: Step 1: Modeling the formation of the hydrophobicity image of the insulator: The modeling process is expressed as: B=I*k+ε (1) Where: k is the blur kernel, I is the potential map, * is the franchise operation, ε is the random noise in the map, and B is the hydrophobic image collected by the imaging system; Step 2: The blur kernel of the high-frequency image is estimated to restore the edge of the insulator hydrophobic image B The fuzzy kernel estimation modeling formula is expressed as: Where: y represents the hydrophobic image The horizontal and vertical gradients of x represent the clear image. The horizontal and vertical gradients of x; k represents the blur kernel, ||·||0 represents the l0 norm, and its function is to calculate the number of non-zero values ​​in x; The solution of blur kernel is carried out in a multi-scale solution framework; the solution steps at each scale can be divided into blur kernel estimation and sharp edge recovery, which can be expressed as: In the blur kernel estimation stage, the blur kernel can be obtained by formula (3); since the blur kernel estimation is performed on the high-frequency part of the image, x represents the horizontal and vertical gradient maps of the potential sharp image, and y represents the horizontal and vertical gradient maps of the blurred image; the formula can be expressed as: Expand the formula to: Solving formula (5) we can get the minimum value of k, as shown in formula (6): In the step of restoring the sharp edges of the hydrophobic image, the terms related to k are removed, and formula (7) is obtained: Step 3: Introduce the auxiliary variable w and transform formula (7) into an equivalent problem in formula (8): Then we get the Lagrangian function, as shown in formula (9), where J1 and J2 represent Lagrangian multipliers and u is a penalty parameter: Therefore, the corresponding alternative iteration scheme is: The minimum value of can be converted into the problem shown in Formula 11: The solution to this problem is as follows: Step 4: In the deconvolution image blurring algorithm; the improved model is shown in formula (13): Where k = exp(-|γ| 0.8 ),γ is defined as: Where B is the hydrophobic image of the insulator, N n (p) is a h*h window centered at each pixel; for small-scale edge information, due to the gradient image The sign of , in the calculation process, is used to eliminate the small-scale edges that are not conducive to the blur kernel estimation.

2. The insulator hydrophobicity image edge detection algorithm in the CPS system taking into account the deconvolution deblurring algorithm according to claim 1, characterized in that: In step 2, formula (2) contains three terms; the first term is the likelihood term, that is, the convolution of the restored clear hydrophobic image x and the blur kernel k is similar to the observed hydrophobic map y; the second term is the l2-norm regularization term applied to the blur kernel, which is used to stabilize the estimation of the blur kernel; the third term is the l0-norm regularization term.

3. The insulator hydrophobicity image edge detection algorithm in a CPS system taking into account the deconvolution deblurring algorithm according to claim 1, characterized in that: In step 2, the convolution operation in the formula is transformed into the multiplication of corresponding points in the frequency domain using the fast Fourier transform, and the derivative of k is calculated to solve formula (5) to obtain the minimum value of k.

4. The insulator hydrophobicity image edge detection algorithm in a CPS system taking into account the deconvolution deblurring algorithm according to claim 1, characterized in that: Formula (7) is a non-convex problem with non-differentiable points. The alternating direction method is used to solve the problem in formula (7).

5. The insulator hydrophobicity image edge detection algorithm in a CPS system taking into account the deconvolution deblurring algorithm according to claim 1, characterized in that: In step 4, in the deconvolution image blur algorithm, large-scale gradient information that has a positive contribution to the blur kernel estimation is screened, and small image gradient information that has a negative contribution to the blur kernel estimation is eliminated.

Images