A reflective portable imaging and image processing method for metal grid defect detection

Abstract

The present invention discloses a reflective portable imaging and image processing method for metal mesh defect detection. The method comprises the following steps: constructing a reflective portable imaging system and combining it with a carrying motion device to capture images of the metal mesh in situ; extracting prior information about block defects and broken wire defects from the metal mesh image using a method based on spectral filtering fusion and Hough transform; inputting the captured image and the prior defect information into a low-rank decomposition model to obtain the low-rank portion, sparse portion, and noise portion of the image; and performing threshold segmentation on the sparse portion to extract defects. The method can achieve in-situ imaging measurement of the metal mesh, identify and detect various defects in the metal mesh image, and provide accurate and efficient detection results.

Description

Technical Field

[0001] The present invention relates to the field of microstructure defect detection, and in particular to a reflective portable imaging and image processing method for metal grid defect detection. Background Art

[0002] The widespread use of electromagnetic equipment has led to problems such as electromagnetic pollution, electromagnetic interference, and electromagnetic information security. Metal mesh is a transparent electromagnetic shielding film that transmits high-frequency visible light and blocks low-frequency microwaves, making it widely used in the field of electromagnetic shielding.

[0003] However, metal grids are prone to defects such as impurity adhesion, wire breakage, and foreign matter during fabrication due to various factors, including process equipment, materials, and human factors. In actual use, metal grid structures can also be damaged by external factors such as scratches, erosion, and impact. These defects can seriously affect the optoelectronic performance of the metal grid.

[0004] Traditionally, metal mesh defect detection is primarily performed through visual observation under a microscope. This method is time-consuming and labor-intensive, with results varying from person to person and susceptible to uncertainties such as the inspector's experience and energy. Manual inspection is insensitive to subtle variations in the metal mesh structure and can only detect defects qualitatively, failing to quantitatively analyze them. This makes it of limited value in guiding improvements to metal mesh manufacturing processes. Furthermore, this method is not suitable for in-situ defect detection of metal meshes.

[0005] Metal mesh defect detection falls under the category of microstructure surface defect detection. Currently, there are relatively few patents regarding metal mesh defect detection methods. Patent 202210829319.X discloses a metal mesh defect detection method based on a structural contrast information stack. This patent first calculates the period of the image, then uses the period size to segment the image into blocks with different starting points. The four-neighborhood structural similarity difference values ​​of the image blocks are calculated to obtain a multi-layer difference matrix, which is then superimposed to form a priori map. A low-rank decomposition model is then used to generate a low-rank, sparse, and noisy image. A posterior mask is constructed using the priori map, and the sparse map is filtered to obtain a saliency map. This is then combined with threshold segmentation to achieve metal mesh defect detection. The defect priori information extraction method provided by this invention suffers from inaccurate information extraction and excessive extraction time, resulting in poor metal mesh defect detection results and low efficiency. Patent 201410131635.5 discloses a method for detecting and identifying metal mesh defects. This patent analyzes the metal mesh structural design information in a GDS II file, simulates metal mesh defects, and builds a defect signature library. This library then trains a support vector machine to develop a metal mesh defect classifier. This classifier is then used to classify metal mesh defects after preprocessing and feature extraction. The simulated defects in this invention deviate from the actual defects, and detection accuracy depends on the size of the defect signature library.

[0006] Patent 202210988249.2 discloses a method for extracting the period of a periodic structure and a wafer defect detection method. This patent applies a Fourier transform to the image under test to obtain a spectrum. Peak-finding the spectrum yields an estimated spatial frequency corresponding to the period. An upsampled Fourier transform is then performed on the image under test. Peak-finding the locally upsampled spectrum yields the precise value of the period. Based on the period, the median image of L adjacent target images is obtained and subtracted from the target image for defect detection and classification. The detection accuracy of this invention is affected by the accuracy of the period. While defect detection through subtraction can improve computational efficiency, it suffers from low accuracy. Patent 202211191900.X discloses a wafer inspection method and its application. This patent obtains a perfect grain image from multiple grain images. A pixel-by-pixel comparison is performed between the image under test and the perfect grain image. A two-dimensional result map is generated based on the difference values, a defect control threshold is set, and defects are marked. The defect detection performance of this invention is affected by the perfect grain image and the defect control threshold, resulting in poor applicability. The perfect grain image must be recalculated for grains of varying sizes.

[0007] Patent 202210363820.1 discloses a visual inspection method for surface defects in microstructures used in intelligent manufacturing. This patent establishes a training set of surface defects on the workpiece to be inspected, builds a YOLOv3-based deep learning model to complete the training, extracts large-scale features of the defects, and transmits the output to the machine learning stage. Small-scale features are further extracted using methods such as SIFT, ORB, and histograms for defect detection. The defect recognition rate of this invention relies on a large training set of defect images. Too few training sets can easily lead to overfitting, and the output is a defect label, failing to capture the specific shape and size of the defect. Patent 202211028774.6 discloses a wafer defect detection and localization algorithm based on a cascaded YOLO-GAN. This patent integrates an improved joint system of object detection, speech segmentation, and image generation. Taking a raw wafer image as input, it uses an improved YOLOv5 object detection model to obtain the detection box position, a BiseNet semantic segmentation model to obtain the wafer foreground mask, and an improved production adversarial network model to reconstruct the wafer image and locate the defect area. Using the position of the target detection frame as a constraint, the defect connectivity domain is analyzed and a Softmax classifier is introduced to achieve defect location and wafer defect segmentation. The defect detection effect of this invention also depends on the quality and quantity of the image dataset and cannot be adapted to detection scenarios where datasets are difficult to collect.

[0008] In summary, in order to complete the quality inspection tasks in the in-situ state during the processing and preparation of metal meshes and the engineering application process, and to quickly and accurately detect the defects of metal meshes, it is necessary to develop a reflective portable imaging and image processing method for metal mesh defect detection. Summary of the Invention

[0009] In order to complete the quality inspection task of metal mesh, while taking into account the measurement requirements, quality and efficiency of defect detection of the metal mesh in its in-situ state, the present invention proposes a reflective portable imaging system in the field of industrial product inspection, which can obtain images of the metal mesh in its in-situ state. Based on the low-rank decomposition model, a priori information extraction method is introduced for square grid and circular grid respectively, and an image processing method for metal mesh defect detection is proposed, which can realize fast and accurate detection of defects in the metal mesh in its in-situ state.

[0010] The technical solutions of the present invention are as follows:

[0011] 1. An image processing method for metal grid defect detection, comprising the following steps:

[0012] Step 1: Capture defect images of the metal grid in its original state;

[0013] Step 2: For the grid-type mesh, use spectrum filtering fusion to extract block defect prior information, and simultaneously extract disconnection defect prior information based on the Hough transform method. The obtained block defect prior information and disconnection defect prior information are superimposed as the defect prior information of the metal mesh image. For the circular mesh, similarly use spectrum filtering fusion to extract block defect prior information, and use the Hough transform and spectrum filtering method to extract disconnection defect prior information. The obtained block defect prior information and disconnection defect prior information are superimposed as the defect prior information of the metal mesh image.

[0014] The spectrum filtering fusion method extracts the prior information of block defects by performing Fourier transform on the input metal grid image to obtain a spectrum graph, filtering the spectrum graph using three filters of increasing size and then performing inverse Fourier transform to obtain filtered images I1, I2, and I3, selecting appropriate thresholds k1, k2, and k3, and performing image fusion using the following formula:

[0015] P1=k1×I3+k2×(I1+I2)-k3×(I2-I1)

[0016] Select an appropriate threshold, set the non-block defect area to zero, and obtain prior information on block defects;

[0017] The method based on Hough transform for extracting prior information of broken wires comprises the following steps: binarizing an input metal grid image, performing Hough transform on the binarized image, detecting straight lines and connecting broken wires to obtain an image I4, performing an expansion operation on the binarized image to obtain I5, performing a difference operation on I4 and I5, selecting a suitable threshold, setting the non-broken wire defect area to zero, and obtaining prior information of the broken wire defect;

[0018] Step 3: The collected metal mesh image and defect prior information are used as inputs of the low-rank decomposition model to perform low-rank decomposition on the image to obtain the low-rank part, sparse part, and noise part of the image;

[0019] Step 4: Perform threshold segmentation on the sparse part obtained by low-rank decomposition to obtain the metal mesh defect detection result.

[0020] 2. In the aforementioned image processing method for metal grid defect detection, in step 1, a reflective portable imaging system is constructed to capture defect images of the metal grid in situ. The reflective portable imaging system and its optical path structure are as follows:

[0021] The reflective portable imaging system consists of an external optical component and a smartphone. The external optical component is coupled with the smartphone camera module to realize the smartphone microscopic imaging function. The optical path structure of the reflective portable imaging system is composed of two superimposed 4f optical path structures. A beam splitter is introduced in the first 4f optical path structure to reflect the illumination light emitted by the light source and transmit the imaging light reflected by the metal grid. An aperture stop is introduced between the two 4f optical path structures to reduce the impact of stray light on imaging quality.

[0022] The step of collecting images of the metal grid in the in-situ state comprises:

[0023] A portable reflective imaging system is fixed on a supporting motion device. By controlling the movement of the supporting motion device, images of different positions of the metal mesh in its in-situ state are scanned and collected, and the smartphone transmits the collected image information to the computer.

[0024] 3. In the above-mentioned image processing method for metal mesh defect detection, in step 2, the step of extracting the prior information of broken wire defects based on the method of Hough transform and spectral filtering is as follows: spectral filtering is performed on the collected image, and a suitable threshold is selected for binarization. This part can be performed simultaneously when the spectral filtering fusion method is used to extract the prior information of block defects, so the impact on the efficiency of the algorithm is extremely low; at the same time, median filtering is performed on the collected image, and the Hough transform is used to detect circular rings and connect broken wires, and a differential operation is performed with the expanded image, and a suitable threshold is selected for binarization; the two results are superimposed as the prior information of the circular ring mesh broken wire defect.

[0025] 4. In the above-mentioned image processing method for metal grid defect detection, in step 3, the low-rank decomposition model is as follows:

[0026]

[0027] Where D represents the input metal mesh image, A represents the low-rank part, which is the periodic structure of the metal mesh that satisfies the low-rank property, E represents the sparse part, which is the defect part that destroys the periodicity of the metal mesh, and N represents the noise part, which is the noise in the metal mesh image. W = exp(-P) is the defect weight matrix, P is the prior information of the defect, which is used to guide the correct decomposition of the sparse part. Parameters λ and β are penalty weights used to balance the decomposition of the low-rank part, the sparse part, and the noise part. * represents the nuclear norm of the matrix, ||·||1 represents the 1-norm of the matrix, ||·|| F Represents the F-norm of the matrix.

[0028] 5. In the above-mentioned image processing method for metal mesh defect detection, in step 4, the threshold segmentation step is: by analyzing the grayscale value of the sparse defect area after low-rank decomposition, the grayscale value of the broken line defect is less than 0, and the grayscale value of the block defect is greater than 0. Due to the large difference in absolute values, a double threshold segmentation method is used to extract the broken line defects and block defects respectively.

[0029] The present invention has the following advantages and outstanding effects:

[0030] 1. The present invention is based on the need for in-situ metal mesh defect detection. It combines a portable microscopic imaging device with a robotic arm to capture metal mesh images as input for a defect detection algorithm. To achieve rapid and accurate metal mesh defect detection, the present invention optimizes a low-rank decomposition algorithm based on the low-rank nature of periodic structures in metal mesh images, while defects that disrupt periodicity exhibit sparsity. This algorithm combines threshold segmentation to identify and detect defects, addressing the poor performance of traditional low-rank decomposition algorithms for detecting large-area block defects and multi-periodic disconnections.

[0031] 2. This invention proposes a reflective portable imaging system for industrial inspection. Laboratory-grade microscopes are not suitable for imaging measurements in industrial inspection, while traditional smartphone microscopes are mostly transmissive and have a short working distance, making it difficult to achieve in-situ imaging measurements of industrial products. This reflective portable imaging system, designed by coupling external optical components with a smartphone camera module, achieves reflective microscopic imaging, increases the system's working distance, and enables in-situ microscopic imaging measurements of industrial products.

[0032] 3. The present invention proposes an algorithm for extracting defect prior information based on spectral filtering fusion and Hough transform. Traditional prior information extraction algorithms suffer from problems such as inaccurate prior information extraction and excessive extraction time, which seriously affect the quality and efficiency of metal mesh defect detection. Based on the characteristics of metal mesh defects, spectral filtering fusion and Hough transform are used to extract prior information of block defects and broken wire defects, respectively. The metal mesh prior information extracted by the present invention is more accurate and more efficient, solving the problem of missed detection and false detection in low-rank decomposition algorithms and improving algorithm efficiency.

[0033] In summary, the present invention uses a reflective portable imaging system to capture images of the metal mesh as input to the defect detection algorithm, thus solving the problem of difficulty in achieving in-situ imaging measurement of the metal mesh. The defect prior information is extracted using a priori information extraction method based on spectral filtering fusion and Hough transform, effectively improving the quality and extraction efficiency of the defect prior information. The defect prior information extracted by the present invention is used to guide low-rank decomposition, decomposing the metal mesh image into a low-rank part, a sparse part, and a noise part, and performing threshold segmentation on the sparse part to extract defects. The invention can quickly and accurately detect metal mesh defects, solving the problem of missed detection and false detection that are prone to occur when using traditional prior information. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of the defect detection method in the present invention.

[0035] Figure 2 Schematic diagram of the optical path structure of the reflective portable imaging system in the present invention.

[0036] Figure 3 Schematic diagram of extracting prior information using the spectrum filtering fusion method in the present invention.

[0037] Figure 4 This is a schematic diagram of extracting prior information using the Hough transform method in the present invention.

[0038] Figure 5 Schematic diagram of extracting prior information using the Hough transform and spectrum filtering method in the present invention.

[0039] Figure 6 Schematic diagram of the low-rank decomposition process in the present invention.

[0040] Figure 7 Schematic diagram of threshold segmentation in the present invention.

[0041] Figure 8 The figure shows the detection results of metal grid images (single-cycle disconnection defects and block defects) using the method of the present invention. DETAILED DESCRIPTION

[0042] The present invention proposes a reflective portable imaging and image processing method for metal grid defect detection. The method of the present invention is described in detail step by step in conjunction with embodiments and drawings.

[0043] This embodiment is a reflective portable imaging and image processing method for metal grid defect detection, and its flow chart is as follows: Figure 1 The method consists of the following steps:

[0044] Step 1: Capture defect images of the metal grid in its original state;

[0045] Step 2: For the grid-type mesh, use spectrum filtering fusion to extract block defect prior information, and simultaneously extract disconnection defect prior information based on the Hough transform method. The obtained block defect prior information and disconnection defect prior information are superimposed as the defect prior information of the metal mesh image. For the circular mesh, similarly use spectrum filtering fusion to extract block defect prior information, and use the Hough transform and spectrum filtering method to extract disconnection defect prior information. The obtained block defect prior information and disconnection defect prior information are superimposed as the defect prior information of the metal mesh image.

[0046] The spectrum filtering fusion method extracts the prior information of block defects by performing Fourier transform on the input metal grid image to obtain a spectrum graph, filtering the spectrum graph using three filters of increasing size and then performing inverse Fourier transform to obtain filtered images I1, I2, and I3, selecting appropriate thresholds k1, k2, and k3, and performing image fusion using the following formula:

[0047] P1=k1×I3+k2×(I1+I2)-k3×(I2-I1)

[0048] Select an appropriate threshold and set the non-blocky defect area to zero to obtain the prior information of blocky defects;

[0049] The method based on Hough transform for extracting prior information of broken wires comprises the following steps: binarizing an input metal grid image, performing Hough transform on the binarized image, detecting straight lines and connecting broken wires to obtain an image I4, performing an expansion operation on the binarized image to obtain I5, performing a difference operation on I4 and I5, selecting a suitable threshold, setting the non-broken wire defect area to zero, and obtaining prior information of the broken wire defect;

[0050] Step 3: Substitute the input image and prior information into the low-rank decomposition model to decompose the metal mesh image into a low-rank part, a sparse part, and a noise part;

[0051] Step 4: Perform threshold segmentation on the sparse part to extract defects;

[0052] Specifically, in step 1, a reflective portable imaging system is constructed to capture defect images of the metal grid in situ. The optical path structure of the reflective portable imaging system is as follows: Figure 2 Reference Figure 2The imaging optical path of a reflective portable imaging system consists of two superimposed 4f optical path structures. The objective lens and the tube lens form the first 4f optical path structure. A large focal length lens is used, and a beam splitter can be placed between the two to reflect the illumination light from the light source and transmit the imaging light reflected by the metal mesh, thereby increasing the working distance of the imaging system. The relay lens and the smartphone's built-in lens group form the second 4f optical path structure. The focal length of the relay lens must be smaller than the focal length of the smartphone's built-in lens group. An aperture stop is introduced between the two 4f optical path structures to reduce the impact of stray light on image quality.

[0053] Specifically, in step 2, the prior information of the grid-type block defect is extracted based on the spectrum filtering fusion method, such as Figure 3 Reference Figure 3 ,The spectrum of the metal mesh image is obtained by Fourier transform, and three square filters of different sizes are designed for filtering, cutting off the high-frequency information of the spectrum and retaining the low-frequency information. The inverse Fourier transform is performed, and the three filtered results are fused. The appropriate threshold is selected to set the non-defect area to zero to obtain the prior of block defects.

[0054] Specifically, in step 2, the prior information of the grid wire break defect is extracted based on the Hough transform method as follows: Figure 4 Reference Figure 4 , binarize the collected image, use the Hough transform method to detect straight lines in the binary image and connect the broken lines, dilate the binary image, perform difference operation on the two, select a suitable threshold to set the non-defective area to zero, and obtain the prior of the broken line defect.

[0055] Specifically, in step 2, the prior information of the ring-shaped grid wire break defect is extracted based on the Hough transform method and spectrum filtering method. Figure 5 Reference Figure 5 , perform spectral filtering on the collected image, and select a suitable threshold for binarization. This part can be performed simultaneously when the spectral filtering fusion method is used to extract the prior information of block defects, so the impact on the efficiency of the algorithm is extremely low; at the same time, perform median filtering on the collected image, use Hough transform to detect circular rings and connect broken lines, perform differential operation with the expanded image, and select a suitable threshold for binarization; the two results are superimposed as the prior information of circular ring-shaped grid broken line defects.

[0056] Specifically, in step 3, the low-rank decomposition process is as follows Figure 6 Reference Figure 6 The prior information extraction method proposed in the present invention is used to extract the prior information of metal grid defects, guide low-rank decomposition, and decompose the image into a low-rank part, a sparse part, and a noise part. The low-rank decomposition effect is better.

[0057] The low-rank decomposition model is as follows:

[0058]

[0059] Where D represents the input metal mesh image, A represents the low-rank part, which is the periodic structure of the metal mesh that satisfies the low-rank property, E represents the sparse part, which is the defect part that destroys the periodicity of the metal mesh, and N represents the noise part, which is the noise in the metal mesh image. W = exp(-P) is the defect weight matrix, P is the prior information of the defect, which is used to guide the correct decomposition of the sparse part. Parameters λ and β are penalty weights used to balance the decomposition of the low-rank part, the sparse part, and the noise part. * represents the nuclear norm of the matrix, ||·||1 represents the 1-norm of the matrix, ||·|| F Represents the F-norm of the matrix.

[0060] Specifically, in step 4, the threshold segmentation process is as follows Figure 7 Reference Figure 7 After low-rank decomposition, the grayscale value of the interrupted line defect in the sparse part is less than 0, and the grayscale value of the block defect is greater than 0. Due to the large difference in absolute values, the double threshold segmentation method is used to extract them separately. In the result figure, the grayscale value of 1 represents the defect area, and the grayscale value of 0 represents the non-defect area.

[0061] The method proposed in this invention is used to detect defects in the metal grid in situ. The detection results are as follows: Figure 8 . Figure 8 The image shown contains single-cycle disconnection defects and block defects. The defect detection results are shown on the right, and the detection results are accurate. Using the method proposed in this invention, it is possible to quickly and accurately detect multiple types of defects in the metal grid in situ.

[0062] The above description is only a specific example of the present invention. It is obvious that for professionals in this field, after understanding the content and principles of the present invention, they may make various modifications and changes in form and details without departing from the concept of the present invention, or directly / indirectly apply it to other related technical fields while still falling within the scope of protection of the claims of the present invention.

Claims

1. An image processing method for metal grid defect detection, characterized in that: The steps include: Step 1: Capture defect images of the metal grid in its original state; Step 2: For the grid-type mesh, the spectrum filtering fusion method is used to extract the block defect prior information, and the Hough transform-based method is used to extract the line break defect prior information. The obtained block defect prior information and the line break defect prior information are superimposed as the defect prior information of the metal mesh image. For the ring-type mesh, the spectrum filtering fusion method is also used to extract the block defect prior information, and the Hough transform-based and spectrum filtering-based method is used to extract the line break defect prior information. The obtained block defect prior information and the line break defect prior information are superimposed as the defect prior information of the metal mesh image. The spectrum filtering fusion method extracts the prior information of block defects by performing Fourier transform on the input metal grid image to obtain a spectrum graph, filtering the spectrum graph using three filters of increasing size and then performing inverse Fourier transform to obtain filtered images I1, I2, and I3, selecting appropriate thresholds k1, k2, and k3, and performing image fusion using the following formula: P1=k1×I3+k2×(I1+I2)-k3×(I2-I1) Select an appropriate threshold and set the non-blocky defect area to zero to obtain the prior information of blocky defects; The method based on Hough transform for extracting prior information of broken wires comprises the following steps: binarizing an input metal grid image, performing Hough transform on the binarized image, detecting straight lines and connecting broken wires to obtain an image I4, performing an expansion operation on the binarized image to obtain I5, performing a difference operation on I4 and I5, selecting a suitable threshold, setting the non-broken wire defect area to zero, and obtaining prior information of the broken wire defect; Step 3: The collected metal mesh image and defect prior information are used as inputs of the low-rank decomposition model to perform low-rank decomposition on the image to obtain the low-rank part, sparse part, and noise part of the image; Step 4: Perform threshold segmentation on the sparse part obtained by low-rank decomposition to obtain the metal mesh defect detection result.

2. The image processing method for metal grid defect detection according to claim 1, characterized in that: In step 1, a reflective portable imaging system is constructed to capture defect images of the metal grid in situ: The reflective portable imaging system and its optical path structure are as follows: The reflective portable imaging system consists of an external optical component and a smartphone. The external optical component is coupled with the smartphone camera module to realize the smartphone microscopic imaging function. The optical path structure of the reflective portable imaging system is composed of two superimposed 4f optical path structures. A beam splitter is introduced in the first 4f optical path structure to reflect the illumination light emitted by the light source and transmit the imaging light reflected by the metal grid. An aperture stop is introduced between the two 4f optical path structures to reduce the impact of stray light on imaging quality. The step of collecting images of the metal grid in the in-situ state comprises: A portable reflective imaging system is fixed on a supporting motion device. By controlling the movement of the supporting motion device, images of different positions of the metal mesh in its in-situ state are scanned and collected, and the smartphone transmits the collected image information to the computer.

3. The image processing method for metal grid defect detection according to claim 1, wherein in step 2: The method based on Hough transform and spectrum filtering for extracting prior information of broken wire defects comprises the following steps: performing spectrum filtering on the collected annular grid image, selecting a suitable threshold for binarization, which can be performed simultaneously when extracting prior information of block defects using the spectrum filtering fusion method, and thus having minimal impact on algorithm efficiency; simultaneously performing median filtering on the collected image, detecting annular rings and connecting broken wires using Hough transform, performing a differential operation with the expanded image, and selecting a suitable threshold for binarization; and superimposing the two results as prior information of the annular grid broken wire defect.

4. The image processing method for metal grid defect detection according to claim 1, characterized in that: In step 3: The low-rank decomposition model is as follows: Where D represents the input metal mesh image, A represents the low-rank part, which is the periodic structure of the metal mesh that satisfies the low-rank property, E represents the sparse part, which is the defect part that destroys the periodicity of the metal mesh, and N represents the noise part, which is the noise in the metal mesh image. W = exp(-P) is the defect weight matrix, P is the prior information of the defect, which is used to guide the correct decomposition of the sparse part. Parameters λ and β are penalty weights used to balance the decomposition of the low-rank part, the sparse part, and the noise part. * represents the nuclear norm of the matrix, ||·||1 represents the 1-norm of the matrix, ||·|| F Represents the F-norm of the matrix.

5. The image processing method for metal grid defect detection according to claim 1, characterized in that: In step 4: The steps of threshold segmentation of the sparse part are: By analyzing the grayscale values ​​of the sparse defect areas after low-rank decomposition, the grayscale value of the broken wire defect is less than 0, and the grayscale value of the block defect is greater than 0. Due to the large difference in absolute values, the double threshold segmentation method is used to extract the broken wire defect and block defect respectively.

Images