An image processing method for high-voltage cable defects

Abstract

The present invention belongs to the technical field of image processing, and particularly relates to an image processing method for high-voltage cable defects. First, the present invention performs processing such as multi-scale Gaussian smoothing on the real-time image of the cable surface collected to solve uneven illumination, enhance the brightness dynamic range and local details; then uses residual learning and frequency-domain high-frequency enhancement to improve the image resolution and details; then performs multi-scale edge detection on the image to be detected to obtain edge features; finally, optimizes the edge details through adaptive edge enhancement and noise suppression. The present invention effectively improves the accuracy and reliability of high-voltage cable defect detection, has significant advantages in image processing efficiency, accuracy and stability, and meets the requirements of high-voltage cable production for surface defect image processing.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and particularly relates to an image processing method for high-voltage cable defects. Background Art

[0002] In modern power systems, high-voltage cables play a crucial role, and their safe and stable operation is essential for the reliability of power transmission. With the development of technology, image processing technology has received extensive attention and application in the field of high-voltage cable defect detection. In the process of high-voltage cable defect detection, acquiring images of the cable surface through various image acquisition devices is the basic step. However, the actual acquisition environment is complex and variable, resulting in many problems in the acquired images. Uneven illumination is one of the common challenges. Uneven illumination can make some areas of the image too bright or too dark, reducing the contrast and clarity of the image, making it difficult to clearly present the defect features on the cable surface, and increasing the difficulty of subsequent defect recognition and analysis. In addition, the resolution of the original acquired images may be limited and unable to meet the requirements for precise detection of tiny defects. Some minor damages on the cable surface, such as tiny scratches and abrasions, may be ignored due to insufficient resolution, thus posing a safety hazard. At the same time, image edge details are crucial for accurately judging the shape, size, and location of defects, but in the existing technology, the extraction and enhancement effects of edge features are often unsatisfactory, possibly resulting in blurred edges or loss of important edge information. Summary of the Invention

[0003] The present invention aims at the technical problems existing in the image processing of high-voltage cable defects and proposes an image processing method for high-voltage cable defects.

[0004] To achieve the above object, the technical solution adopted by the present invention is as follows: including the following steps: including the following steps:

[0005] S1. Collect real-time images of the cable surface to generate an image to be detected; process the problem of uneven illumination in the image to be detected to enhance the brightness dynamic range and local details. The specific implementation steps are as follows:

[0006] S11. First, perform multi-scale Gaussian smoothing on the input image I(x, y) to separate the reflection component and the illumination component: Wherein, R IMSR (x, y) is the pixel value of the image after improved multi-scale Retix processing, γ i is the contrast enhancement parameter for each size, ω i is the multi-scale weight, satisfying G(x, y, σ i ) is a two-dimensional Gaussian kernel function, and the scale parameter is σ i ;

[0007] S12. Secondly, based on the adaptive illumination distribution compensation, enhance the uniformity of light to obtain the compensated image where L(x, y) is the large-scale illumination distribution and β is the global brightness adjustment coefficient;

[0008] S13. Finally, fuse R IMSR (x, y) and I DLCM (x, y) to compress the dynamic range: where max(R IMSR (x, y), I DLCM (x, y)) is the maximum value of the pixel values of R IMSR (x, y) and I DLCM (x, y) at the corresponding pixel positions;

[0009] S2. Then, it is necessary to process the pixel values of the image processed in S1. Through the improved residual mapping and frequency-domain adaptive high-frequency enhancement technology, improve the image resolution and details;

[0010] S3. Then, perform multi-scale edge detection on the image to be detected to obtain edge features at different scales, specifically: where E scale (x, y) represents the multi-scale edge response, is the Laplacian edge response at scale k, and ω k is the adaptive weight at different scales, which is used to control the contribution degree of each scale to the total edge response;

[0011] S4. Finally, use the adaptive edge enhancement technology to further optimize the multi-scale edges to enhance the edge details, and apply noise suppression processing. The formula is: E enhanced (x, y) = [E scale (x, y) + γ·G(x, y)]·(1 + α·N(x, y)), where γ is the coefficient that controls the intensity of the image gradient G(x, y), α is the enhancement factor, and N(x, y) represents the local noise estimation of the image, and the processing of the detected image is completed.

[0012] Preferably, the specific implementation of improving the image resolution and details in step S2 is as follows:

[0013] S21. First, perform residual learning on the input low-resolution image I low (x, y) to obtain the high-resolution residual I res (x, y): I res (x, y) = I pred (x, y) - I low (x, y), where I pred(x, y) is the target high-resolution image predicted by the neural network, and the high resolution is obtained by adding the low resolution and the residual: I RRM (x, y) = I low (x, y) + I res (x, y), where I RRM (x, y) is the high-resolution image after residual mapping;

[0014] S22. Then, perform frequency-domain high-frequency enhancement on the residual result, enhance the high-frequency components in the frequency domain through Fourier transform, and improve the edge details of the image. Finally, obtain the frequency-domain enhanced image I freq (x, y) = F -1 [F high (u, v)], where F -1 represents the inverse Fourier transform, (u, v) are the coordinates in the frequency space, and there is a corresponding relationship with the spatial coordinates (x, y) of the image;

[0015] S23. Finally, fuse the residual mapping image and the frequency-domain enhanced image through weights to obtain the high-resolution output image I final (x, y) = α q ·I RRM (x, y) + (1 - α q )·I freq (x, y), where α q is the weight coefficient, and its value range is from 0 to 1.

[0016] Preferably, the specific implementation of step S22 is as follows:

[0017] S221. First, perform Fourier transform on the image to obtain the frequency-domain representation F RRM (u, v);

[0018] S222. Use an adaptive high-frequency filter to enhance the high-frequency components of the frequency-domain image: where is high-frequency filtering, σ H is the range of high-frequency enhancement, which is dynamically adjusted according to the size and features of the image;

[0019] S223. Finally, convert the enhanced frequency-domain image back to the spatial domain to obtain the frequency-domain enhanced image: I freq (x, y) = F -1 [F high (u, v)].

[0020] Preferably, the Laplacian operator in step S3 is used for extracting edge features at each scale, and adaptively adjusts the edge response at each scale through the following formula: Where δ is the adjustment coefficient, which controls the sensitivity of weight change, and μ is the edge enhancement mean, which enables the algorithm to adapt to different defect scales.

[0021] Preferably, the edge enhancement processing in step S4 is implemented by adding a gradient enhancement term γ·G(x,y) to enhance edge contrast, and stability is achieved by local noise suppression N(x,y). The noise estimation formula is: Where p and q are window sizes, controlling the local extent of the noise estimate.

[0022] Compared with the prior art, the advantages and positive effects of the present invention are that in the treatment of uneven illumination, multi-scale Gaussian smoothing and other technologies are used, fine processing is performed according to different scales, adaptive illumination distribution compensation is performed for reasonable dimming, and fusion compression is performed to ensure suitability, thereby overcoming the defects of traditional means, improving image contrast and clarity, and making defect features clearly presented. For resolution improvement, residual learning and frequency domain high-frequency enhancement technology are used. Residual learning uses low-resolution information, and frequency domain high-frequency enhancement is used for precise processing, avoiding the problem that traditional methods cannot take into account edge details or introduce artifacts, and can accurately detect tiny defects. In terms of edge detection, multi-scale edge detection combined with adaptive edge enhancement and noise suppression technology performs well, can accurately extract edges of defects of different scales, and solves the problems of blurred edges, insensitivity to tiny defects, and susceptibility to noise interference of existing methods. It has obvious advantages in efficiency, accuracy and stability, and meets the requirements of surface defect image processing in high-voltage cable production. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 It is a schematic diagram of the method flow of the present invention; DETAILED DESCRIPTION

[0025] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0027] Embodiment. In the operation and maintenance of high-voltage cables, their safe and stable operation is of great significance to power transmission. Traditional detection methods have gradually revealed many deficiencies when faced with complex actual working conditions. For example, manual inspection is difficult to accurately detect subtle defects and is inefficient. With the development of technology, image processing technology has been applied in cable defect detection, but still faces challenges. The collected cable images are often affected by uneven illumination. For example, in cable tunnels, the light distribution is complex, resulting in some areas of the image being too bright or too dark, making it difficult to distinguish defect features. At the same time, the resolution of the original image is limited, and small defects are easily overlooked. The extraction and enhancement of edge features are not good, affecting the judgment of the shape, size, and position of defects. To overcome these problems, the present invention aims to provide an effective image processing method to improve the accuracy and reliability of high-voltage cable defect detection and ensure the safe operation of the power system. The overall process is as Figure 1 shown.

[0028] First, in order to improve the image contrast and clarity, make the defect features on the cable surface clearly presented, and solve the problem that it is difficult to identify defects caused by uneven illumination, the present invention first uses multi-scale Gaussian smoothing on the input image I(x,y) to separate the reflection component and the illumination component, and obtains an image with uniform illumination. The formula is: where, R IMSR (x,y) is the pixel value of the image after improved multi-scale Retix processing, γ i is the contrast enhancement parameter for each size, ω i is the multi-scale weight, satisfying G(x,y,σ i ) is a two-dimensional Gaussian kernel function, and the scale parameter is σ i ; Secondly, based on adaptive illumination distribution compensation to enhance the uniformity of light, the compensated image is obtained, where L(x,y) is the large-scale illumination distribution, and β is the global brightness adjustment coefficient; Finally, R IMSR (x,y) and I DLCM (x,y) are fused to compress the dynamic range: where max(R IMSR (x,y),I DLCM (x,y)) is the maximum value of the image after illumination uniformity; Compared with traditional illumination processing methods, it is difficult to adapt to complex illumination, is vulnerable to local details, and makes it difficult to distinguish defects. The present invention uses technologies such as multi-scale Gaussian smoothing. Multi-scale Gaussian smoothing is finely processed according to different scales, adaptive illumination distribution compensation reasonably adjusts the light, and fusion compression ensures suitability.

[0029] In order to achieve the purpose of accurately detecting minute defects and meeting the requirements for detecting subtle damages on the cable surface, the present invention will next process the image resolution and details. When traditional methods improve the resolution, they often fail to balance edge details or introduce artifacts. The residual learning and frequency-domain high-frequency enhancement technology of the present invention utilize low-resolution information in residual learning and precisely process it in frequency-domain high-frequency enhancement. The specific process is as follows: First, perform residual learning on the input low-resolution image I low (x, y) to obtain the high-resolution residual I res (x, y): I res (x, y) = I pred (x, y) - I low (x, y), where I pred (x, y) is the target high-resolution image predicted by the neural network. The high resolution is obtained by adding the low resolution and the residual: I RRM (x, y) = I low (x, y) + I res (x, y), where I RRM (x, y) is the high-resolution image after residual mapping; then, perform frequency-domain high-frequency enhancement on the residual result. Enhance the high-frequency components in the frequency domain through Fourier transform. Perform Fourier transform on the image to obtain the frequency-domain representation F RRM (u, v); Use an adaptive high-frequency filter to enhance the high-frequency components of the frequency-domain image: where is high-frequency filtering, σ H is the range of high-frequency enhancement, which is dynamically adjusted according to the size and features of the image; Finally, convert the enhanced frequency-domain image back to the spatial domain to obtain the frequency-domain enhanced image: I freq (x, y) = F -1 [F high (u, v)], where F -1 represents the inverse Fourier transform, (u, v) are the coordinates in the frequency space, which have a corresponding relationship with the spatial coordinates (x, y) of the image; Finally, fuse the residual mapping image and the frequency-domain enhanced image through weights to obtain the high-resolution output image I final (x, y) = α q ·I RRM (x, y) + (1 - α q )·I freq (x, y), where α q is the weight coefficient, and its value range is from 0 to 1.

[0030] Finally, in order to accurately obtain edge features at different scales and enhance edge details to better judge defect features, a multi-scale edge detection and optimization technology is adopted. Perform multi-scale edge detection on the image to be detected to obtain edge features at different scales. Specifically: where Escale (x,y) represents the multi-scale edge response, is the Laplace edge response at scale k, ω k are adaptive weights of different scales, used to control the contribution of each scale to the total edge response. Where δ is the adjustment coefficient, which controls the sensitivity of weight change, and μ is the mean value of edge enhancement. Finally, the adaptive edge enhancement technology is used to further optimize the multi-scale edges to enhance edge details, and noise suppression is applied. The formula is: E enhanced (x,y)=[E scale (x,y)+γ·G(x,y)]·(1+α·N(x,y)), where γ is the coefficient that controls the intensity of the image gradient G(x,y), α is the enhancement factor, and N(x,y) represents the local noise estimation of the image. The processing of the detection image is completed. Where p and q are window sizes, which control the local range of noise estimation. Compared with existing edge detection methods, when faced with complex textures and defects of different scales on the cable surface, there may be problems such as blurred edges, insensitivity to minor defects, and susceptibility to noise interference. The multi-scale edge detection of the present invention combines adaptive edge enhancement and noise suppression technology to perform well. Multi-scale edge detection uses the Laplace operator and adaptive weights to accurately extract the edges of defects of different scales. Compared with existing methods, this method has significant advantages in image processing efficiency, accuracy and stability, and meets the requirements of surface defect image processing in high-voltage cable production.

[0031] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An image processing method for high-voltage cable defects, characterized in that, It includes the following steps: S1. Collect the real-time image on the cable surface to generate the image to be detected; process the uneven illumination problem in the image for the image to be detected to enhance the brightness dynamic range and local details. The specific implementation steps are as follows: S11. First, perform multi-scale Gaussian smoothing on the input image I(x, y) to separate the reflection component and the illumination component: Among them, R IMSR (x, y) is the pixel value of the image after improved multi-scale Retix processing, and γ i is the contrast enhancement parameter for each size, and ω i is the multi-scale weight, satisfying G(x, y, σ i ) is a two-dimensional Gaussian kernel function with the scale parameter σ i ; S12. Secondly, based on the adaptive illumination distribution compensation, enhance the uniformity of light to obtain the compensated image where L(x, y) is the large-scale illumination distribution and β is the global brightness adjustment coefficient; S13. Finally, fuse R IMSR (x, y) and I DLCM (x, y) to compress the dynamic range: where max(R IMSR (x,y), I DLCM (x,y)) is the maximum value of the pixel values of R IMSR (x,y) and I DLCM (x,y) at the corresponding pixel positions; S2. Then, it is necessary to process the pixel values of the image obtained by S1 processing, and improve the image resolution and details through the improved residual mapping and frequency-domain adaptive high-frequency enhancement technology; S3. Next, perform multi-scale edge detection on the image obtained by processing S2 to obtain edge features at different scales, specifically: where E scale (x, y) represents the multi-scale edge response, is the Laplacian edge response at scale k, and ω k is the adaptive weight for different scales, which is used to control the contribution degree of each scale to the total edge response; S4. Finally, use the adaptive edge enhancement technology to further optimize the multi-scale edges to enhance edge details, and apply noise suppression processing. The formula is: E enhanced (x, y) = [E scale (x, y) + γ · G(x, y)] · (1 + α · N(x, y)), where γ is the coefficient controlling the intensity of the image gradient G(x, y), α is the enhancement factor, and N(x, y) represents the local noise estimate of the image, thus completing the processing of the detected image.

2. The image processing method for high-voltage cable defects according to claim 1, characterized in that The specific implementation of step S2 to improve the image resolution and details is as follows: S21. First, perform residual learning on the input low-resolution image to obtain the high-resolution image after residual mapping; S22. Next, perform frequency-domain high-frequency enhancement on the residual result, enhance the high-frequency components in the frequency domain through Fourier transform, and improve the edge details of the image. Finally, obtain the frequency-domain enhanced image I freq (x,y) = F -1 [F high (u,v)], where F -1 represents the inverse Fourier transform, (u,v) are the coordinates in the frequency space, and there is a corresponding relationship with the spatial coordinates (x,y) of the image; Finally, the high-resolution output image I is obtained by fusing the residual mapping image and the frequency-domain enhanced image with weights. final (x,y) = α q ·I RRM (x,y) + (1 - α q )·I freq (x,y), where α q is the weight coefficient, and its value range is from 0 to 1. Among them, the specific implementation of step S22 is as follows: S221. First, perform a Fourier transform on the image to obtain the frequency-domain representation F RRM (u, v); S222. Enhance the high-frequency components of the frequency-domain image using an adaptive high-frequency filter: where is high-frequency filtering, and σ H is the range of high-frequency enhancement, which is dynamically adjusted according to the size and features of the image; Finally, convert the enhanced frequency-domain image back to the spatial domain to obtain the frequency-domain enhanced image: I freq (x,y) = F -1 [F high (u,v)].

3. A method for processing images of high-voltage cable defects according to claim 1, characterized in that, The Laplacian operator in step S3 is used for extracting edge features at each scale, and adaptively adjusts the edge responses at each scale through the following formula: where δ is an adjustment coefficient that controls the sensitivity of the weight change, and μ is the edge enhancement mean, enabling the algorithm to adapt to different defect scales.

4. A method for image processing of high-voltage cable defects according to claim 1, characterized in that In the step S4, the edge enhancement processing is realized by adding a gradient strengthening term γ·G(x,y) to enhance the edge contrast, and the stability is realized by local noise suppression N(x,y). The noise estimation formula is as follows: where p and q are the window sizes, controlling the local range of noise estimation.

Images