Method for eliminating scratch defect in cross section of high-voltage cable

Abstract

The present invention relates to a method for eliminating scratch defects in the cross section of a high-voltage cable, comprising: first performing morphological processing on an input image to obtain a new modality; and then calculating a multimodal fused image, and using the fused image as the input of a total variation algorithm to finally eliminate scratch defects in the cross section of the high-voltage cable. Provided in the present invention is a novel improved total variation algorithm, and by introducing multimodal features, the problem of image information loss caused by a staircasing effect in the traditional total variation algorithm is relieved. The present invention can retain as much information of an original image as possible in the process of eliminating scratch defects, thereby remarkably improving the detection accuracy.

Description

A method for eliminating scratch defects in the cross section of a high-voltage cable Technical Field

[0001] The invention relates to a method for eliminating scratch defects on the cross section of a high-voltage cable, in particular to a method for eliminating scratch defects on the cross section of a high-voltage cable based on an improved total variation algorithm. Background Art

[0002] As the main carrier of power transmission, high-voltage cables are an important part of the power distribution system. Therefore, ensuring the production quality of high-voltage cables is of great significance. High-voltage cables are generally composed of core conductors, insulation layers, shielding layers, and protective layers. Therefore, the quality of high-voltage cables is generally evaluated by detecting the length of the cable, the thickness of the insulation layer, and the number of core conductors. Among them, the number of conductors in the cable is one of the most important indicators for measuring cable quality. At present, an efficient method is to detect the number of conductors on the cross-sectional image of the cable through machine vision. However, during the cable cutting and sampling process, many scratches are generated, which seriously affects the quality of the conductor number detection. Therefore, it is necessary to design a method that can effectively eliminate scratch-type defects on the cross section.

[0003] This paper considers the scratch defects generated during the cable cutting process as noise in the image and uses denoising methods to eliminate the scratches. Numerous algorithms have been studied for image denoising, which can be divided into two main categories: deep learning-based image denoising algorithms and traditional image denoising algorithms.

[0004] Since the introduction of convolutional neural networks (CNNs), and especially after the ImageNet 2012 challenge, deep learning-based defect detection and denoising algorithms have become a popular approach in the denoising field. While these algorithms can achieve better performance than previous ones, they require large amounts of data for model training. However, due to the high cost of high-voltage cables, the available data is relatively limited. Furthermore, scratch defects generated during the cutting process are distributed in a chaotic manner, making them insufficient for model training.

[0005] Traditional denoising algorithms are mainly categorized into spatial domain denoising, transform domain denoising, and model-based denoising. Spatial domain denoising refers to the correlation between image pixels in the spatial domain and is primarily achieved through filters. Classic filtering denoising methods include median filtering, mean filtering, Wiener filtering, and Gaussian filtering, such as the vector median filter (VMF), bilateral filtering, and the TSM (Tri-state Median) nonlinear filter. These filters can effectively preserve image edges and details, but they do not consider global features. Transform domain denoising separates image information from noise through various operations, denoises the image based on the characteristics of the noise, and then performs an inverse transform to return to the original image. Common transform domain denoising algorithms are based on Fourier transforms (FT), discrete cosine transforms (DCT), and wavelet transforms (WT). However, transform domain denoising suffers from high computational complexity and threshold dependence. Model-based denoising algorithms model the distribution of image noise and use the model distribution as a priori to try to obtain clear images and optimization algorithms. For example, the total variation (TV) algorithm proposed by Rudin et al. is a general model that applies partial differential equations to image denoising. The main idea of ​​this model is to reduce the sum of the gradient integrals in the image in the pixel domain. This model allows sharp discontinuous points in the image, but it is prone to staircase effects, which cause texture disappearance and make the image blurred. Summary of the Invention

[0006] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a method for eliminating scratch defects in the cross section of a high-voltage cable based on an improved total variation algorithm.

[0007] The technical solution adopted in the present invention is:

[0008] A method for eliminating scratch defects in the cross section of a high-voltage cable is proposed. The method first performs morphological processing on the input image to obtain a new modality. Then, a fused image of multiple modalities is calculated and used as the input of a total variation algorithm to ultimately eliminate scratch defects in the high-voltage cable.

[0009] Preferably, the image is first dilated by a morphological processing method, and then the dilated image is eroded; wherein the structure element used is defined as follows:

[0010] (1)

[0011] Among them, I A is the input image, I B is a morphologically processed image, where S1 and S2 are the structural elements used in the dilation and erosion processes, respectively; It is the result of dilation and erosion of the input image;

[0012] Then the images of different modalities are fused to obtain a new image I in :

[0013] (2)

[0014] α and β are different modal weights, 0≤α≤1, 0≤β≤1, α+β=1.

[0015] As a preferred method, the improved total variation algorithm first uses equations (1) and (2) to obtain the new image I in , then I in As the input of Equation (3), the initial value of the output image I is set to the original image I A ; Finally, perform iterative update to obtain the output image I* with scratch defects eliminated:

[0016] (3)

[0017] Ω represents the pixel domain of the entire image, and λ represents the regularization parameter; For authenticity, is a TV regularization term that integrates multiple modalities; x, y represent the horizontal and vertical pixel coordinate positions of the image, dx, dy represent the horizontal and vertical gradients, I(x, y), Represent the grayscale and gradient vector of coordinates (x, y) respectively;

[0018] Given that

[0019] (4)

[0020] The loss function F is defined as follows:

[0021] (5)

[0022] where F is abbreviation of;

[0023] In order to find the minimum value of function F, equation (6) is used to obtain equation (7) of function F:

[0024] (6)

[0025] (7)

[0026] Therefore, the optimized output image I of the kth iteration is k+1 for:

[0027] (8)

[0028] I k 、 Represents the image grayscale and gradient of the kth iteration.

[0029] As a preference, a dataset of high-voltage cable cross-section images was built. The dataset contains high-voltage cable cross-section images of several standard sizes, 200 images of each size, for a total of 1,000 images.

[0030] Preferably, the sizes include five standard sizes: 240mm, 300mm, 400mm, 800mm and 1000mm.

[0031] The beneficial effects of the present invention are:

[0032] The present invention proposes a novel and improved total variation algorithm, which alleviates the problem of image information loss caused by the staircase effect in the traditional total variation algorithm by introducing multimodal features; the present invention can retain as much information of the original image as possible in the process of eliminating scratch-type defects, and can significantly improve detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] FIG1 is a diagram of a high-voltage cable cross-section acquisition device according to the present invention;

[0034] FIG2 is a partial image of the dataset of the present invention;

[0035] FIG3 is a diagram showing the effect of eliminating scratch defects in the cross section of a high-voltage cable according to the present invention;

[0036] FIG4 is a diagram showing the effect of detecting the number of conductors according to the present invention;

[0037] FIG5 is an image analysis diagram showing the effect of the present invention on eliminating scratch-type defects in the cross section of a high-voltage cable. DETAILED DESCRIPTION

[0038] The technical solutions of the present invention are further specifically described below through examples. These examples are provided for the purpose of illustrating the present invention and are not intended to limit the present invention. All other examples obtained by persons of ordinary skill in the art based on the examples in this application without creative work are intended to fall within the scope of protection of this application. Example 1

[0039] A method for eliminating scratch defects in the cross section of high-voltage cables first performs morphological processing on the input image to obtain a new modality. Then, a fused image of multiple modalities is calculated and used as the input of an improved total variation algorithm to ultimately eliminate scratch defects in the high-voltage cable.

[0040] Multimodal fusion

[0041] Each image source or form can be called a modality. Different modalities exhibit different representations and information, leading to overlapping and complementary information. Therefore, by fusing information from multiple modalities, we can enrich image features.

[0042] To better preserve image edge information, this paper proposes a unique multimodal fusion method. This method primarily applies different processing techniques to the same image, such as sparse processing, filtering, and morphological processing, to generate different modalities. After analyzing and comparing these methods, morphological processing was ultimately selected.

[0043] The morphological processing method is to first dilate the image and then erode the dilated image. This method eliminates small holes between objects and fills the gaps in the contours. It is similar to the closing operation in morphology, but it uses a different structuring element, which is defined as follows:

[0044] (1)

[0045] Among them, I A is the input image, I B is a morphologically processed image, where S1 and S2 are the structural elements used in the dilation and erosion processes, respectively; It is the result of dilation and erosion of the input image.

[0046] Then the images of different modalities are fused to obtain a new image I in :

[0047] (2)

[0048] α and β are different modal weights, 0≤α≤1, 0≤β≤1, α+β=1.

[0049] Improved TV algorithm

[0050] The present invention proposes an improved total variation algorithm, which utilizes the redundancy and complementarity of multimodal fusion to enrich image information, reduce the staircase effect, and better preserve the edge information of the image.

[0051] The algorithm of the present invention takes the multimodal fused image as the algorithm input, and then processes it with the total variation algorithm, ultimately achieving the elimination of scratch-type defects in the cross section of the high-voltage cable.

[0052] The algorithm first uses equations (1) and (2) to obtain the new image I in , then I in As the input of Equation (3), the initial value of the output image I1 is set to the original image I in ; Finally, perform iterative update to obtain the output image I* with scratch defects eliminated:

[0053] (3)

[0054] Ω represents the pixel domain of the entire image, and λ represents the regularization parameter; For authenticity, is a TV regularization term that integrates multiple modalities; x, y represent the horizontal and vertical pixel coordinate positions of the image, dx, dy represent the horizontal and vertical gradients, I(x, y), Represents the grayscale and gradient vector of coordinates (x, y) respectively.

[0055] Given that

[0056] (4)

[0057] The loss function F is defined as follows:

[0058] (5)

[0059] where F is abbreviation of;

[0060] In order to find the minimum value of function F, equation (6) is used to obtain equation (7) of function F:

[0061] (6)

[0062] (7)

[0063] Therefore, the optimized output image I of the kth iteration is k+1for:

[0064] (8)

[0065] I k 、 Represents the image grayscale and gradient of the kth iteration. Example 2

[0066] Refer to Figure 1 for a self-built experimental platform for capturing cross-sectional images of high-voltage cables. The platform primarily consists of a light source controller 1, an industrial camera 2, a camera fixture 3, an optical lens 4, an experimental template support 5, a ring light source 6, a high-voltage cable 7, and a tablet computer. The tablet computer runs the Windows 10 64-bit operating system.

[0067] Using the aforementioned experimental platform, we built a dataset of high-voltage cable cross-section images. This dataset contains 200 images of high-voltage cable cross-sections in five standard sizes: 240 mm, 300 mm, 400 mm, 800 mm, and 1000 mm, for a total of 1000 images. Some images from this dataset are shown in Figure 2.

[0068] The following experiments use a self-developed dataset and compare the improved total variation algorithm (Method A) proposed in this paper with the improved frequency-domain filtering method (Method B) proposed in the literature (Beiping, H.; Xiaogang, Z.; Wen, Z.; Tianliang, C.; Lingchao, C. Research on texture removal of the cable core image based on frequency domain filtering. Chinese Journal of Scientific Instrument 2021, 42) and the traditional total variation algorithm (Method C). All experiments were performed using an Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz and 8.00 GB of memory, and the software environment was MATLAB R2020a.

[0069] 1. Experiment on the removal effect of scratch defects on the cross section of high-voltage cables

[0070] In order to verify the performance of the method proposed in this invention (method A), method A was compared with method B on a self-built data set. The evaluation index used for each specification in the experiment was the average elimination rate, which is defined as follows:

[0071] (9)

[0072] Herev p It represents the average scratch elimination rate of the p-th specification high-voltage cable, n p represents the number of images of the p-th cable, M p,i The total number of scratches in the i-th image of the p-th specification cable, m p,i N represents the number of scratches eliminated in the i-th image of the p-th specification cable. p is the total number of cables of each specification, and p is the total number of cable specifications.

[0073] In this experiment, P=5, N 1,2…,P = 200. Figure 3 shows the effect of removing scratch-type defects from a high-voltage cable cross section (partial experimental visualization). (a) shows the cross section of a high-voltage cable with a scratch-type defect, (b) shows the removal result using Method A, and (c) shows the removal result using Method B. Comparing the experimental results in Figure 3, we can see that Method B can remove most shallow scratches, but is less effective for deeper scratches. Furthermore, the removal effect is not particularly ideal for high-voltage cables with a large number of conductors and complex scratch-type defect distribution. These issues are effectively addressed by Method A, as shown in Figure 3, which can remove most scratch-type defects.

[0074] To further analyze the scratch removal effect, the removal rate of each sample was calculated. The results of the scratch defect removal rates of Method A and Method B for high-voltage cables of different specifications are shown in Table 1.

[0075] Table 1

[0076]

[0077] Table 1 shows that Method B achieved a scratch removal rate approaching 50% in only some cable specifications, while Method A, proposed in the present invention, achieved a scratch removal rate exceeding 90% in all specifications. Compared to Method B, Method A, proposed in the present invention, eliminated nearly all scratches in some samples. Visual image analysis and data analysis demonstrate that the proposed method is more effective in removing scratch-type defects.

[0078] 2. Conductor quantity detection experiment

[0079] To verify the effectiveness of method A in improving the detection of the number of high-voltage cable conductors, we used the Hough circle detection algorithm on our own dataset and conducted the following experiments: 1. After processing the original image using method A, Hough circle detection was performed on the output image. 2. After processing the original image using method B, Hough circle detection was performed on the output image. 3. Hough circle detection was performed directly on the original image (direct detection). The evaluation metric used in the experiment was the average detection rate of the number of conductors, which is defined as follows:

[0080] (10)

[0081] Among them, v d,p N represents the average detection rate of p types of conductors. p is the total number of cables of each specification, D p,i represents the total number of conductors in the i-th image of the p-th specification cable, d p,i Indicates the number of conductors detected in the i-th image of the p-th specification cable.

[0082] In this experiment, P=5, N 1,2…,p =200, the conductor quantity detection effect diagram (part of the experimental visualization effect) is shown in Figure 4, where group (a) shows the detection effect after using method A, group (b) shows the detection effect after using method B, and group (c) shows the effect of direct detection. From Figure 4, it can be seen intuitively that due to the presence of some deeper scratches that directly divide the conductor into two, the detection of the number of conductors is affected, resulting in poor detection effect of direct detection. The reason why some conductors are not detected in method B is that only shallow scratches are eliminated in the cable sample of the present invention, and all scratches are not eliminated. Therefore, the actual number of conductors is interfered by some scratches that have not been eliminated. Compared with the first two methods, method A can eliminate most of the scratches while retaining the edge information of the image, thereby improving the quality of conductor quantity detection.

[0083] The results of the average detection rate of the number of conductors by different methods are shown in Table 2:

[0084] Table 2

[0085]

[0086] As can be seen from Table 2, the method of the present invention (Method A) significantly improves the performance of conductor quantity detection. Specifically, the average detection rate is improved by about 30%. The improvement in average detection rate brought by the method of the present invention is 1.5 times that of Method B.

[0087] 3. Ablation Experiment

[0088] [Corrected 17.07.2024 according to Rule 26] In order to further illustrate the effect of the present invention, this experiment selected the traditional total variation algorithm (method C) for comparison, and used the image quality evaluation index and running time as the evaluation criteria for defect elimination. Among them, the image quality evaluation index mainly selected is the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). The specific elimination effect, that is, the image analysis of the effect of eliminating scratch-type defects in the cross section of the high-voltage cable is shown in Figure 5. From left to right, it represents samples 1-5. Group (a) is the cross section of the high-voltage cable with scratch-type defects; Group (b) is the elimination result using the method of the present invention (method A); Figure (c) is the elimination result using the traditional total variation algorithm. By comparing the elimination effect diagram of Figure 5, it is found that under the same experimental data and parameters, both methods can eliminate most of the scratches. However, there are still some scratches that cannot be eliminated in method C.

[0089] The comparative data results of the two methods for eliminating scratch defects are shown in Table 3.

[0090] Table 3

[0091]

[0092] Table 3 shows that Method A achieves significantly better peak signal-to-noise ratio and structural similarity than Method C. Furthermore, Method A's peak signal-to-noise ratio is consistently above 0.75, demonstrating that Method A effectively preserves useful image information while eliminating scratch-type defects. Although Method A performs an additional multimodal operation compared to Method C, its overall speed is virtually unchanged. This demonstrates that the present invention significantly improves algorithm performance.

[0093] This paper collects cross-sectional images of high-voltage cables of varying specifications to create a dataset, with each image displaying a varying number and pattern of scratch defects. Experimental results demonstrate that this method can effectively eliminate different types of scratch defects in high-voltage cable cross-sections, significantly improving the accuracy of conductor count detection.

[0094] Since many scratches will be left on the cross section during the cutting process of high-voltage cables, these scratches will affect the detection of the number of conductors. Therefore, the present invention proposes an improved total variation algorithm to eliminate these scratch defects. The algorithm first performs morphological processing on the input image to obtain a new modality; then calculates the fusion image of multiple modalities, and uses the fused image as the input of the total variation algorithm, and finally eliminates the scratch defects of the high-voltage cable. The experimental results show that the method proposed by the present invention not only achieves a scratch elimination rate of more than 90%, but also improves the average detection rate of the number of conductors by 30%. Moreover, since the algorithm introduces multimodal features, it can eliminate scratches that cannot be eliminated in the traditional total variation algorithm, and the peak signal-to-noise ratio and structural similarity are both good, which alleviates the problem of image information loss caused by the staircase effect in the traditional total variation algorithm.

[0095] Therefore, the method proposed in the present invention can significantly improve the accuracy of conductor quantity detection, making quality monitoring during the high-voltage cable production process more effective.

[0096] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for eliminating scratch defects in the cross section of a high-voltage cable, characterized by: First, the input image is morphologically processed to obtain a new modality; then the fused image of multiple modalities is calculated, and the fused image is used as the input of the total variation algorithm to finally eliminate the scratch defects of the high-voltage cable.

2. The method for eliminating scratch defects in a high-voltage cable cross section according to claim 1, characterized in that: The image is first dilated by morphological processing, and then the dilated image is eroded; the structure element used is defined as follows: （1） Among them, I A is the input image, I B is a morphologically processed image, where S1 and S2 are the structural elements used in the dilation and erosion processes, respectively; It is the result of dilation and erosion of the input image; Then the images of different modalities are fused to obtain a new image I in : （2） α and β are different modal weights, 0≤α≤1, 0≤β≤1, α+β=1.

3. The method for eliminating scratch defects in a high-voltage cable cross section according to claim 2, characterized in that: The improved total variation algorithm first uses equations (1) and (2) to obtain the new image I in , then I in As the input of Equation (3), the initial value of the output image I is set to the original image I A ; Finally, perform iterative update to obtain the output image I* with scratch defects eliminated: （3） Ω represents the pixel domain of the entire image, and λ represents the regularization parameter; For authenticity, is a TV regularization term that integrates multiple modalities; x, y represent the horizontal and vertical pixel coordinate positions of the image, dx, dy represent the horizontal and vertical gradients, I(x, y), Represent the grayscale and gradient vector of coordinates (x, y) respectively; Given that （4） The loss function F is defined as follows: （5） where F is abbreviation of; In order to find the minimum value of function F, equation (6) is used to obtain equation (7) of function F: （6） （7） Therefore, the optimized output image I of the kth iteration is k+1 for: （8） I k 、 Represents the image grayscale and gradient of the kth iteration.

4. The method for eliminating scratch defects in a high-voltage cable cross section according to claim 1, characterized in that: A dataset of high-voltage cable cross-section images was built. The dataset contains 1,000 images of high-voltage cable cross-sections in several standard sizes, 200 images of each size.

5. The method for eliminating scratch defects in a high-voltage cable cross section according to claim 4, characterized in that: The sizes include five standard sizes: 240mm, 300mm, 400mm, 800mm and 1000mm.

Images