Method and system for detecting surface defects of an insulation layer of a polyethylene insulated cable

Abstract

The present application relates to the technical field of image processing, more particularly, the present application relates to a kind of polyethylene insulated cable's insulating layer surface defect detection method and system, comprising: the surface image of the insulating layer is obtained;The defect response index of each pixel in surface image is calculated, the product of the gradient amplitude of each pixel in the corresponding local window is positively correlated with the defect response index, and the absolute value of the difference of the gray mean value of all pixels in the local window is positively correlated with the defect response index.The present application constructs defect response index by introducing gradient direction angle variance, uses the physical characteristics of the consistent gradient direction of highlight area and the disorder of defect area to distinguish the two, and constructs defect saliency enhancement factor accordingly to dynamically adjust Gaussian kernel scale.The method automatically uses large kernel smoothing interference in highlight area, and uses small kernel to retain details in defect area, effectively overcome the limitations of traditional fixed parameter algorithm, and significantly improve the defect detection precision under complex reflective background.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for detecting surface defects in the insulation layer of polyethylene insulated cables. Background Technology

[0002] As a core component of power transmission systems, the integrity and surface quality of polyethylene insulated cables directly determine their electrical performance, operational safety, and service life. In modern cable manufacturing processes, rigorous surface defect inspection is typically performed on cables produced on the production line to ensure product quality.

[0003] In actual testing scenarios, cables on production lines are in a state of high-speed continuous movement, and the polyethylene insulation layer has a smooth cylindrical structure with strong reflective properties. When industrial cameras are used in conjunction with lighting equipment for image acquisition, high-brightness specular reflection bands (highlight areas) inevitably form on the cable surface. This highlight phenomenon not only occupies a prominent position in the image but also often obscures the true surface texture.

[0004] However, existing surface defect detection techniques, such as traditional level set-based image segmentation methods (e.g., the LBF model), typically rely on globally fixed kernel scale parameters for evolution. When processing cable images that exhibit both strong specular interference and subtle defect features, this fixed-parameter approach faces an irreconcilable practical contradiction: if a larger kernel scale is chosen to suppress interference and background noise in specular areas, the edges of extremely small scratches or shallow bubbles on the insulation surface may be excessively smoothed, leading to missed detections; conversely, if a smaller kernel scale is chosen to capture these subtle defect details, the algorithm is highly susceptible to overreacting to specular edges and noise, mistakenly identifying specular reflection bands as defects. Summary of the Invention

[0005] This invention provides a method and system for detecting surface defects in the insulation layer of polyethylene insulated cables, aiming to solve the problem in related technologies of how to accurately separate weak defects from a complex background while suppressing high-light interference.

[0006] In a first aspect, the present invention provides a method for detecting surface defects in the insulation layer of a polyethylene insulated cable, comprising: acquiring a surface image of the insulation layer; calculating a defect response index for each pixel in the surface image, wherein the defect response index is positively correlated with the product of the gradient magnitudes of each pixel within a corresponding local window, positively correlated with the absolute value of the difference between the gray-level mean values ​​of all pixels within the local window, and positively correlated with the variance of the gradient direction angles of all pixels within the local window; constructing a defect saliency enhancement factor for each pixel, wherein the defect saliency enhancement factor is positively correlated with the absolute value of the difference between the defect response index and the global mean, and positively correlated with the absolute value of the difference between the defect response index and the global minimum; using the defect saliency enhancement factor as a control parameter to construct an adaptive kernel scale; wherein the adaptive kernel scale is negatively correlated with the defect saliency enhancement factor; and segmenting the preprocessed image using a local binary fitting model including the adaptive kernel scale, and extracting regions within the contour as surface defects. By constructing a defect response index that integrates features such as gradient direction angle variance, the system effectively distinguishes between specular regions (with consistent gradient directions) and real defects (with chaotic gradient directions). Based on this, an adaptive kernel-scale LBF model is established, which solves the contradiction between suppressing specular interference and preserving subtle defect details in existing fixed kernel-scale techniques. This enables accurate segmentation and detection of surface defects such as tiny scratches and bubbles under strong reflective backgrounds.

[0007] Furthermore, the local window corresponding to each pixel is obtained, including: for any pixel, constructing a window of size W centered on that pixel as the local window corresponding to that pixel. This clarifies the spatial range of local feature calculation, ensuring that the grayscale and gradient statistics within the pixel neighborhood can be fully utilized when calculating the defect response index, providing a unified and effective calculation benchmark for accurately measuring the degree of texture oscillation in local areas.

[0008] Furthermore, regions within the contour are extracted as surface defects, including: obtaining the Gaussian kernel scale parameters corresponding to each pixel in the entire image and substituting them into... The model iterates through the energy functional, continuously updating the level set function as the number of iterations increases until the contour curve no longer changes significantly or reaches the preset maximum number of iterations. Finally, the zero level set contour of the level set function is extracted, and the internal region enclosed by this contour represents the surface defects of the insulation layer. Utilizing the level set evolution mechanism of the Local Binary Fitting (LBF) model, the contour curve automatically shrinks and closely fits the true edge of the defect under the constraint of adaptive kernel scale. This achieves automated closed segmentation of irregularly shaped defect regions on the surface of the insulation layer, extracting accurate defect contours from the preprocessed image without manual intervention.

[0009] Furthermore, the formula for calculating the defect saliency enhancement factor is as follows: In the formula, Image coordinates The defect saliency enhancement factor corresponding to the local window; Image coordinates The defect response index corresponding to the local window; This represents the average of all defect response indices in the entire graph; This represents the minimum value of the response index of all defects in the entire graph; This is a sensitivity adjustment parameter. By utilizing the deviation of the defect response index from the global statistical characteristics (mean and minimum), weak defect signals are doubly amplified, while the response of background areas close to the global mean is greatly suppressed. This significantly widens the numerical difference between the defect area and the background and highlight areas, providing a high-contrast control basis for subsequent kernel-scale spatial adaptive adjustment.

[0010] Furthermore, the value of the sensitivity adjustment parameter is obtained by adaptive threshold segmentation of the set of all defect response indices using Otsu's method. By using Otsu's method to adaptively threshold segment the set of defect response indices to determine the sensitivity adjustment parameter, the algorithm eliminates the need for manual experience to set fixed parameters and can automatically find the optimal segmentation point based on the actual grayscale distribution characteristics of each frame of the image.

[0011] Furthermore, after acquiring the surface image of the insulating layer, the method further includes: preprocessing the surface image, which includes: removing background and noise using morphological opening and closing operations, and smoothing the image using median filtering. Introducing morphological opening and closing operations and median filtering before image segmentation effectively filters out interference from conveyor belt textures in the production background and impulse noise generated during image acquisition. This cleans the image data while preserving the main structural information of the insulating layer surface, reducing the misleading effect of background noise on subsequent gradient calculations and defect identification.

[0012] Furthermore, the preprocessing also includes gamma correction of the surface image to enhance the contrast of the image edge regions. Introducing morphological opening and closing operations and median filtering before image segmentation effectively filters out interference from conveyor belt textures in the production background and impulse noise generated during image acquisition. This cleans the image data while preserving the main structural information of the insulating layer surface, reducing the misleading effect of background noise on subsequent gradient calculations and defect identification.

[0013] Furthermore, acquiring surface images of the insulation layer includes simultaneously capturing images of the insulation layer surface using a multi-channel linear CCD camera arranged around the circumference of the polyethylene cable. The multi-channel linear CCD camera arrangement adapts to the high-speed, continuous movement characteristics of cable production lines, achieving 360-degree full-coverage synchronous imaging of the cylindrical cable insulation layer surface without blind spots. This ensures the integrity of the inspection data and meets the requirements of industrial production lines for online real-time full inspection.

[0014] Furthermore, by using the Sobel operator to calculate the gradient magnitude of each pixel, edge information in the image can be extracted quickly and effectively, accurately reflecting the rate of change of pixel grayscale.

[0015] In a second aspect, a system for detecting surface defects in the insulation layer of a polyethylene insulated cable is also provided, comprising a processor and a memory, the memory storing a computer program, the processor executing the computer program to implement the method for detecting surface defects in the insulation layer of a polyethylene insulated cable as described in any of the preceding claims.

[0016] Beneficial effects: By introducing gradient direction angle variance to construct a defect response index, the physical characteristics of consistent gradient directions in highlight areas and disordered gradient directions in defect areas are utilized to distinguish between the two. Based on this, a defect saliency enhancement factor is constructed to dynamically adjust the Gaussian kernel size. This method automatically uses a large kernel to smooth interference in highlight areas and a small kernel to preserve details in defect areas, effectively overcoming the limitations of traditional fixed-parameter algorithms and significantly improving defect detection accuracy under complex reflective backgrounds. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart illustrating a defect detection method according to an embodiment of the present invention; Figure 2 This is a schematic image of the surface of the insulating layer to be inspected according to an embodiment of the present invention; Figure 3 This is a schematic illustration of a Gaussian kernel-scale feature map according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the detection results according to the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figures 1 to 4 As shown, S101: Image acquisition and preprocessing of the insulating layer surface.

[0020] In this embodiment, given that polyethylene insulated cables typically move at high speeds continuously on the production line and have a cylindrical surface, in order to achieve surface... For comprehensive, blind-spot-free detection, for example, a three-channel approach can be used. High-speed linear array with angled distribution Cameras are deployed synchronously around the circumference of the cable to capture surface images of the insulation layer, which serve as the raw images. To provide a uniform and stable lighting environment for the cable surface, a high-frequency ring shadowless camera can be configured at the acquisition site. Light source. As a preferred solution, a polarizer can be installed outside the light source to filter out some of the specular reflection caused by the smooth properties of the insulating layer material, thus initially reducing highlight interference in the image. Nevertheless, highlight reflection bands will still inevitably exist in the image.

[0021] After acquiring the original image, targeted preprocessing is required. First, morphological opening and closing operations are used to remove interference from the background conveyor belt and noise, extracting the region of interest containing only the surface of the cable insulation layer. Then, considering the brightness attenuation at image edges due to the curvature of the cable surface, gamma correction is used to non-linearly adjust the image brightness, improving the contrast of the edge regions. Finally, to address potential impulse noise in the production environment, a 3×3 window median filter is applied to smooth the image, resulting in the preprocessed image. ,in Represents the spatial coordinates of a pixel.

[0022] S102: Construct the defect response index.

[0023] In this embodiment, to distinguish between defects and highlights at the feature level, this step aims to construct an index that can effectively quantify the irregularity of gray-level changes and gradient dispersion in local areas. The reason is that although highlight reflection bands are bright, they are essentially specular reflections, and their internal gray-level changes are continuous and smoothly transitioned, thus exhibiting a high degree of consistency in their local gradient directions. Conversely, defects (such as scratches and bubbles) disrupt the smooth structure of the insulating layer surface, resulting in diffuse reflection or scattering, causing severe and disordered oscillations in gray-level within small local areas, and consequently, chaotic gradient directions. Based on this, a defect response index is constructed, and the specific construction process is as follows.

[0024] First, define a size as A sliding window to view the preprocessed image. Each pixel in Centered on a specific point, a local window corresponding to that point is captured, where the size... The value is Pixels. Within each window, calculate the gradient magnitude, gradient direction angle, and average gray value of all pixels within that window. The gradient magnitude of each pixel can be calculated using the Sobel operator.

[0025] Then, pixels are constructed according to the following formula. Defect response index at the location : In the formula, Represents image coordinates The defect response index corresponding to the local window; Image coordinates The total number of pixels within the corresponding local window; Representing coordinates The corresponding local window within the first The gradient magnitude of each pixel; Representing coordinates The corresponding local window within the first The grayscale value of each pixel; Image coordinates The average grayscale value of all pixels within a local window centered on the pixel; It is a very small positive number used to prevent the denominator from being zero. In this embodiment, its preferred value is... ; Represents image coordinates It is the variance of the gradient direction angles of all pixels within the local window centered on it.

[0026] Understandably, this formula integrates three dimensions—gradient strength, grayscale deviation, and gradient direction dispersion—through a product. As the formula shows, for highlight regions, although their grayscale values... The gradient is relatively high, but due to the smooth transition of illumination, its gradient direction is highly consistent, resulting in a high variance in the gradient direction. The value is extremely small, thus making the final calculated The value is relatively small. Conversely, for defective regions, due to diffuse reflection caused by surface roughness, the local gradient magnitude is relatively large. The large gradient variance, coupled with its chaotic gradient directions, leads to significant gradient direction variance. A significant increase; the combined effect of these two factors will make The value increases sharply. Therefore, this index can effectively suppress the response in highlight areas while highlighting the texture features of defect areas, providing a highly discriminative feature map for subsequent processing.

[0027] S103: Construct a defect saliency enhancement factor.

[0028] While the defect response index constructed using the above steps can distinguish between highlights and defects, its corresponding texture features may be weak when dealing with extremely small scratches or shallow bubbles on the cable surface. This results in a calculated defect response index value that, although higher than the background area, is still not significant enough and is easily ignored in subsequent segmentation threshold selection. According to Weber's law, the perceived intensity of a stimulus depends not only on its absolute value but also on its ratio to the background intensity. Therefore, it is necessary to evaluate the degree of deviation of the current local index value from the global statistical distribution.

[0029] This step constructs a defect saliency enhancement factor to achieve nonlinear amplification of weak defect signals. The specific construction process is as follows: First, all calculated defect response index values ​​are compiled into a dataset, and this dataset is treated as a global sequence. The global average value of this sequence is then calculated. and global minimum .

[0030] Then, pixels are constructed according to the following formula. Defect saliency enhancement factor : In the formula, Image coordinates The defect saliency enhancement factor corresponding to the local window; Image coordinates The defect response index corresponding to the local window; This represents the average of all defect response indices in the entire graph; This represents the minimum value of the response index of all defects in the entire graph; As a sensitivity adjustment parameter, a preferred approach is to obtain the value of this parameter by adaptive thresholding of the histogram of the dataset using Otsu's method, thereby enabling it to be automatically adjusted according to the feature distribution of the current image.

[0031] As can be seen from the formula, when the defect response index approaches the global mean... When, i.e., the normal background area, the first term of the formula Approaching Meanwhile, the second item It is also relatively small, thus greatly suppressing the response in the background region and effectively reducing background noise. And when Much larger When the defect area is reached, both parameters increase simultaneously, resulting in a double amplification effect on the defect signal. This nonlinear mapping significantly widens the numerical difference between the defect area and the background area, providing a high-contrast basis for the accurate determination of subsequent algorithm parameters.

[0032] S104: Construct adaptive Gaussian kernel scaling parameters.

[0033] In this embodiment, the traditional The model uses a fixed-scale Gaussian kernel function to define the local energy term, which has limitations when handling complex scenes. If the kernel scale is too large, it will blur the edges of small defects, leading to a decrease in segmentation accuracy; if the kernel scale is too small, the model is extremely sensitive to local gradient changes in noise and highlight regions, easily producing incorrect segmentation contours. Ideally, a larger kernel scale should be used in smooth backgrounds and highlight regions to smooth out interference, while a very small kernel scale should be used in defects and their edge regions to accurately capture details. This step transforms the defect saliency enhancement factor into... The key control parameter required by the algorithm is the adaptive adjustment of the standard deviation of the Gaussian kernel.

[0034] Then, pixels are constructed according to the following formula. Adaptive Gaussian kernel scaling parameters at the location : In the formula, For pixels The corresponding Gaussian kernel standard deviation, i.e., the kernel scale; and These are the preset minimum and maximum kernel scale values, which are 2 and 16 respectively; Image coordinates The defect saliency enhancement factor corresponding to the local window; This is a scale contraction coefficient used to control the rate at which the kernel scale changes with significance; in this embodiment, it is taken as a constant. .

[0035] From the logical relationship of the formula, it can be seen that when the pixel When the standard deviation of the corresponding Gaussian kernel is large, it indicates a defect region with strong significance, and the denominator... It will increase accordingly, making It decays rapidly and approaches the minimum value. At this point, the algorithm uses a small-scale kernel, which can precisely fit the complex contours of the defect. When pixels The corresponding Gaussian kernel standard deviation value is small or close to When the area is in the background or highlight region, the denominator tends to be close to , making Approaching the maximum value At this point, the algorithm employs a large-scale kernel, utilizing the averaging effect over a wide range to ignore local interference from highlights and noise. By introducing the Gaussian kernel standard deviation corresponding to each pixel, spatial adaptive adjustment of the kernel scale parameter is successfully achieved.

[0036] S105: LBF model evolution and defect detection based on adaptive kernel scale.

[0037] After obtaining the Gaussian kernel scale parameters for each pixel in the entire image, substitute them into... In the model's energy functional. Traditional The model relies on a globally fixed kernel function, while this invention improves it to a spatially variable kernel function whose standard deviation varies with spatial location.

[0038] The algorithm's execution flow is as follows: First, initialize a closed level set function. ,in During the iterative evolution process, the behavior of the evolution curve (i.e., the zero-level set) is controlled by the locally adaptive kernel scale parameter. In the defect region, because the kernel scale parameter automatically decreases according to the calculation result of step S104, the zero-level set curve can sensitively respond to the gradient information of the defect edge, thus shrinking and closely fitting the true edge of the defect. In the highlight region, due to the kernel scale... With a larger size, the evolution curve will be subject to stronger smoothing constraints and will not be truncated or misled by subtle gradient changes within the highlights.

[0039] As the number of iterations increases, the level set function The process continues until the contour curve no longer changes significantly or the preset maximum number of iterations is reached. Finally, the level set function is extracted. The zero-level set contour is obtained. The internal region enclosed by this contour is the surface defect region of the insulation layer that is finally detected by the method of this invention. Thus, automated and high-precision inspection of the surface quality of polyethylene insulated cables is completed.

[0040] The present invention also provides a system for detecting surface defects in the insulation layer of polyethylene insulated cables. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the method for detecting surface defects in the insulation layer of polyethylene insulated cables according to the first aspect of the present invention.

[0041] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0042] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0043] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for detecting surface defects in the insulation layer of a polyethylene insulated cable, characterized in that, include: Obtain a surface image of the insulating layer and preprocess the surface image; for any pixel, construct a window of size W centered on that pixel as the local window corresponding to that pixel; Calculate the defect response index of each pixel in the surface image, satisfying: ； Represents image coordinates The defect response index corresponding to the local window; Image coordinates The total number of pixels within the corresponding local window; Representing coordinates The corresponding local window within the first Gradient magnitude of each pixel; Representing coordinates The corresponding local window within the first The grayscale value of each pixel; Image coordinates The average grayscale value of all pixels within a local window centered on the pixel; It is a very small positive number, used to prevent the denominator from being zero; Represents image coordinates The variance of the gradient direction angles of all pixels within the local window centered on the pixel. Construct a defect saliency enhancement factor for each pixel that satisfies: ； Image coordinates The defect saliency enhancement factor corresponding to the local window; This is the average value of the response indices of all defects in the entire graph; This represents the minimum value of the response index of all defects in the entire graph; This is a sensitivity adjustment parameter; Using the defect saliency enhancement factor as a control parameter, an adaptive kernel scale is constructed that satisfies: ； For pixels The corresponding Gaussian kernel standard deviation; and These are the preset minimum and maximum kernel scale values, respectively; This is a scaling factor used to control the rate at which the kernel scale changes with significance; A local binary fitting model with adaptive kernel scale is used to segment the preprocessed image, and the region within the contour is extracted as surface defect.

2. The method for detecting surface defects in the insulation layer of a polyethylene insulated cable according to claim 1, characterized in that, Extracting the region within the contour as surface defects includes: After obtaining the Gaussian kernel scale parameters for each pixel in the entire image, substitute them into... The model energy functional is iterated. As the number of iterations increases, the level set function is continuously updated until the profile curve no longer changes significantly or reaches the preset maximum number of iterations. Finally, the zero level set profile of the level set function is extracted. The internal region enclosed by this profile is the surface defect of the insulation layer.

3. The method for detecting surface defects in the insulation layer of a polyethylene insulated cable according to claim 1, characterized in that, The value of the sensitivity adjustment parameter is obtained by adaptive threshold segmentation of the set of all defect response indices using the Otsu method.

4. The method for detecting surface defects in the insulation layer of a polyethylene insulated cable according to claim 1, characterized in that, After acquiring the surface image of the insulating layer, the process further includes: The preprocessing includes: removing background and noise using morphological opening and closing operations, and smoothing the image using median filtering.

5. The method for detecting surface defects in the insulation layer of a polyethylene insulated cable according to claim 4, characterized in that, The preprocessing further includes performing gamma correction on the surface image to enhance the contrast of the image edge regions.

6. The method for detecting surface defects in the insulation layer of a polyethylene insulated cable according to claim 1, characterized in that, Obtain a surface image of the insulating layer, including: A multi-channel linear CCD camera, arranged around the circumference of the polyethylene cable, is used to simultaneously photograph the surface of the insulation layer.

7. The method for detecting surface defects in the insulation layer of a polyethylene insulated cable according to claim 1, characterized in that, The gradient magnitude of each pixel is calculated using the Sobel operator.

8. A system for detecting surface defects in the insulation layer of a polyethylene insulated cable, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the method for detecting surface defects in the insulation layer of polyethylene insulated cables as described in any one of claims 1-7.

Images