A method and system for detecting ribbon defects based on machine vision

Abstract

This invention proposes a machine vision-based webbing defect detection method and system, aiming to address the low efficiency and poor accuracy of traditional manual webbing inspection. The system uses a webbing cutting module to define an inspection area. Dimension detection, character detection, webbing defect detection, and webbing density detection modules collect data, which is then fed into corresponding data analysis modules for processing. The data is then aggregated into a comprehensive webbing data analysis module to assess webbing quality. Webbing defect detection encompasses surface defects, edge protrusions and depressions, and burr detection, using Gaussian low-pass filtering, threshold processing, and morphological operations. By establishing corresponding coefficients for each data analysis model, a comprehensive data analysis model is constructed to derive a webbing quality assessment index, which is then compared with a standard value to determine webbing quality compliance. This method and system effectively improves the efficiency and accuracy of webbing inspection, meeting the quality inspection requirements of modern webbing production.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition and quality inspection, and in particular to a method and system for detecting defects in a ribbon based on machine vision. Background Art

[0002] Product quality is paramount in the webbing manufacturing industry. Traditional webbing inspection methods rely heavily on manual visual inspection, a method with numerous drawbacks. Manual inspection is extremely inefficient, making rapid and comprehensive testing difficult for large-scale webbing production, leading to extended production cycles. Furthermore, the accuracy of manual inspection is significantly affected by factors such as the inspector's experience, fatigue, and focus, making it prone to missed inspections or misjudgments, making it impossible to guarantee consistent product quality.

[0003] With the continuous improvement of market requirements for webbing quality and the continuous expansion of production scale, traditional manual inspection methods can no longer meet modern production needs. Therefore, there is an urgent need for an efficient and comprehensive automated inspection method and system to replace it, so as to ensure the quality of webbing products and improve production efficiency. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and system for detecting defects in a webbing, which can perform efficient and comprehensive automated detection of the webbing, thereby ensuring the quality of the webbing products and improving production efficiency.

[0005] In order to solve the above technical problems, the technical solution of the present invention is: a ribbon defect detection system based on machine vision, comprising the following modules:

[0006] The cropping module is used to remove the background from the image captured by the camera and divide the detection area. Specifically, it includes:

[0007] Convert the color image into grayscale image and perform Gaussian filtering to remove noise;

[0008] The ribbon outline is extracted by threshold segmentation and Canny edge detection, excluding the background area;

[0009] The size detection module is used to detect the width, thickness and length of the ribbon and obtain the actual size data through image stitching and pixel ratio conversion;

[0010] The character detection module is used to detect the fuzzy, offset, missing, incomplete, and deformed characters on the ribbon. Specifically, it includes:

[0011] Grayscale conversion, Canny edge detection, connected region analysis and contour comparison;

[0012] Density detection module, used to measure the average value of the warp and weft density of the ribbon and compare it with the standard value;

[0013] Defect detection module, including surface defect detection, edge protrusion and depression detection, and burr detection, specifically using Gaussian low-pass filtering, threshold processing, morphological operations and connected component analysis;

[0014] The data analysis module group includes the size data analysis module, character data analysis module, density data analysis module and defect data analysis module, which input the inspection data into the corresponding model to generate the size coefficient , character coefficient , density coefficient , and defect coefficient ;

[0015] Comprehensive data analysis module, used to integrate various coefficients to generate ribbon quality evaluation index And compare it with the standard value to determine whether the webbing is qualified;

[0016] The result judgment and information interaction module is used to display the test results and trigger early warning signals based on the quality assessment index.

[0017] As a preferred solution of the present invention, the specific steps of the size detection module include:

[0018] The edge detection results are used to extract the ribbon contour and the width is calculated by the horizontal boundary points. ;

[0019] Calculate the average thickness by the vertical distance between the upper and lower edge points ;

[0020] Measure the total length of the ribbon by stitching multiple images through SURF feature point matching and affine transformation .

[0021] As a preferred solution of the present invention, the defect detection module further includes an adaptive adjustment mechanism, which can dynamically optimize the detection threshold and light source parameters according to the ribbon material, color and historical data.

[0022] As a preferred embodiment of the present invention, the surface defect detection comprises the following steps:

[0023] Filter the image texture in the frequency domain using a Gaussian low-pass filter;

[0024] Global thresholding, morphological operations, and connected component filtering;

[0025] The pixel sum is calculated through edge detection and mask operation to determine the presence of defects.

[0026] As a preferred solution of the present invention, the edge protrusion and depression detection includes the following steps:

[0027] Binarize and morphologically dilate the grayscale image;

[0028] Perform Canny edge detection after removing small connected components;

[0029] The edge images are partitioned and grouped for straight line fitting analysis.

[0030] As a preferred solution of the present invention, the burr detection includes the following steps:

[0031] Generate a mask and define the edge detection area;

[0032] connecting fracture edges through dilation operations;

[0033] Small area connected components are filtered and burr defects are determined by pixel sum threshold.

[0034] As a preferred solution of the present invention, the quality evaluation index of the comprehensive data analysis module is The judgment rules are:

[0035] when The webbing is judged to be qualified when

[0036] when When the warning signal is triggered;

[0037] in is the standard quality index, is the allowable deviation ratio.

[0038] The present invention also discloses a method for detecting webbing defects based on machine vision, comprising the following steps:

[0039] Divide the detection area through the cropping module;

[0040] Perform size, character, density and defect inspections sequentially and collect data;

[0041] Input the data into the corresponding analysis model to generate coefficients;

[0042] Comprehensively evaluate the quality of the webbing by combining various coefficients and output the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram showing the results of surface defect detection;

[0044] Figure 2 A schematic diagram showing the detection results of edge protrusions and depressions;

[0045] Figure 3 A schematic diagram showing the burr detection results;

[0046] Figure 4To reflect the ribbon image cutting module flow chart;

[0047] Figure 5 To reflect the flow chart of the dimension measurement module;

[0048] Figure 6 To reflect the character detection module flow chart;

[0049] Figure 7 To reflect the flow chart of the ribbon defect detection module;

[0050] Figure 8 To reflect the structural diagram of the ribbon defect detection system;

[0051] Figure 9 The figure shows the flow chart of the webbing defect detection method. DETAILED DESCRIPTION

[0052] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings to make the technical solutions of the present invention easier to understand and grasp.

[0053] A machine vision-based webbing defect detection system consists of the following modules, which work together to detect webbing defects:

[0054] 1. Cropping module

[0055] Function: Remove the background from the image captured by the camera and define the webbing detection area.

[0056] Implementation steps:

[0057] S1, convert the color image into a grayscale image and apply a Gaussian filter to remove noise;

[0058] S2, separating the ribbon area from the background by threshold segmentation;

[0059] S3, use the Canny edge detection algorithm to extract the ribbon outline and eliminate background interference;

[0060] S4. Divide the detection area (such as edge, middle) according to the ribbon contour and transmit the area information to the subsequent module.

[0061] 2. Size detection module

[0062] Function: Detect the width of the webbing ,thickness and length .

[0063] Implementation steps:

[0064] S5, width detection: extract the coordinates of the left and right edges of the ribbon and calculate the horizontal distance;

[0065] S6, thickness detection: extract the upper and lower edge coordinates and calculate the mean vertical distance;

[0066] S7, length detection:

[0067] S7-1, matching local feature points of multiple images using SURF algorithm;

[0068] S7-2, using affine transformation for image registration and stitching;

[0069] S7-3. Convert the total length of the ribbon in the stitched image based on the pixel ratio.

[0070] 3. Character detection module

[0071] Function: Detect blur, position shift, missing or incomplete, and deformation problems of ribbon characters.

[0072] Implementation steps:

[0073] S9, converting the image into grayscale and extracting character outlines using Canny edge detection;

[0074] S10, blur detection: by edge intensity mean With threshold Comparative judgment;

[0075] S11. Position offset: Calculate the Euclidean distance between the character's center of gravity and the expected position , exceeding the threshold Then determine the offset;

[0076] S12, missing / incomplete: determined by comparing the area of ​​the connected region with the standard value;

[0077] S13. Deformation: Compare and detect the geometric feature differences between the character outline and the standard outline.

[0078] 4. Density detection module

[0079] Function: measure the mean value of webbing warp and weft density and with standard density contrast.

[0080] Implementation steps:

[0081] S14, sampling density data multiple times at different positions of the webbing;

[0082] S15. Calculate the density mean and transmit it to the data analysis module.

[0083] 5. Defect detection module

[0084] Function: Detect surface defects, edge protrusions and depressions, and burrs.

[0085] Implementation steps:

[0086] Surface defect detection:

[0087] S16, applying a Gaussian low-pass filter in the frequency domain to suppress the high-frequency noise of the ribbon texture;

[0088] S17, generate a binary image by global threshold processing, and optimize the region by combining morphological operations;

[0089] S18, use Canny edge detection and mask operation to calculate the total number of pixels in the defect area. If it exceeds the threshold, it is determined that there is a surface defect. .

[0090] Edge protrusion and depression detection:

[0091] S19, binarizing and morphologically dilating the grayscale image;

[0092] S20, performing Canny edge detection after removing small connected components;

[0093] S21. Divide the edge image into horizontal partitions and analyze the edge straight line fitting deviation in groups .

[0094] Glitch detection:

[0095] S22, generating a mask to limit the edge detection area;

[0096] S23, connecting the broken edges through dilation operations and filtering small-area connected components;

[0097] S24, calculate the sum of the pixels of the final image, and if it exceeds the threshold, it is determined that there is a burr .

[0098] 6. Data analysis module group

[0099] Function: Input the test data into the model to generate the quality coefficient.

[0100] Implementation steps:

[0101] S25, Dimensional data analysis module:

[0102] Calculation of dimensional deviation coefficient ,in, They are length deviation, width deviation and thickness deviation, is the target length, is the allowable error range, is the target width, is the allowable error range, is the target thickness, is the allowable error range. are the weights of length, width, and thickness deviations, respectively;

[0103] S26, Density data analysis module: Calculate density coefficient ,in, is the density coefficient, is the measured density, is the standard density;

[0104] S27, character data analysis module: calculate character coefficient ,in The influence coefficients of character fuzziness, character position offset, missing or incomplete characters, and character deformation on the character quality coefficient are respectively used to detect the fuzziness of ribbon characters, character position offset, missing or incomplete characters, and character deformation. express;

[0105] S28, defect data analysis module: calculate defect coefficient ,in, is the defect coefficient, The influence coefficients of surface defects, edge protrusions and depressions, and burrs on the overall defect coefficient of the ribbon are respectively; the surface defects, edge protrusions and depressions, and burr defect data are marked as Indicates surface defects, Indicates edge protrusions and depressions, Indicates burr defects.

[0106] 7. Comprehensive data analysis module

[0107] Function: Integrate various coefficients to generate ribbon quality evaluation index .

[0108] Implementation steps:

[0109] S29, according to weight calculate ,in, , are the coefficient influence weights of size, density data, character defect data, and defect data respectively;

[0110] S30, comparison standard quality index, Ratio to allowable deviation , :

[0111] like , determine that the ribbon is qualified;

[0112] like , triggering an early warning signal.

[0113] 8. Result judgment and information interaction module

[0114] Function: Visualize detection results and trigger early warning.

[0115] Implementation steps:

[0116] S31, displaying the ribbon size, characters, density and defect detection results;

[0117] S32. Automatically send a pass / fail report to the production management system based on the quality assessment index.

[0118] 9. Adaptive adjustment mechanism

[0119] Function: Dynamically optimize detection parameters based on webbing material, color and historical data.

[0120] Implementation steps:

[0121] During the initial inspection, the color, texture and light transmittance of the ribbon are analyzed;

[0122] Adjust light source intensity, image contrast and detection threshold based on historical data;

[0123] Reduce false detections and missed detections through continuous learning and optimization algorithms.

[0124] The present invention also discloses a method for detecting defects in a webbing, comprising the following steps:

[0125] S100, dividing the detection area by a cropping module;

[0126] S200, sequentially performing size, character, density and defect inspections and collecting data;

[0127] S300, inputting data into the corresponding analysis model to generate coefficients;

[0128] S400: Evaluate the quality of the webbing by comprehensively evaluating various coefficients and output the results.

[0129] The algorithm used by the adaptive adjustment mechanism is as follows:

[0130] In the actual implementation process, the present invention constructs a mathematical mapping relationship based on the analysis of the image frequency domain energy distribution. , using the linear model

[0131] type The filter standard deviation is dynamically adjusted to achieve adaptive optimization of the filter for images with different webbing texture structures, thereby reducing false detection and missed detection.

[0132] The algorithm flow is as follows:

[0133] 1. Input image Perform Fourier transform to obtain the spectrum amplitude diagram;

[0134] 2. Extract the radius of the central area The low-frequency energy , and calculate the total energy , get the high frequency ratio

[0135] 3. According to the linear mapping function Dynamically calculate Gaussian low-pass filter parameters;

[0136] The specific process is as follows:

[0137] 1. Original image spectrum representation:

[0138] The input image is a two-dimensional grayscale image , its discrete Fourier transform is:

[0139] The image spectrum amplitude is:

[0140] 2. Definition of high frequency energy ratio:

[0141] Let the spectrum center be , with it as the center and radius The circular area is defined as the low-frequency area , the total frequency domain is ,but:

[0142] Total Energy:

[0143] Low frequency energy:

[0144] High frequency energy:

[0145] High frequency energy ratio:

[0146] 3. Adaptive Filter Parameter Model

[0147] According to the high frequency ratio , define the Gaussian filter standard deviation parameter The linear model is: in, is the minimum filtering scale when the low-frequency features of the image are obvious, The maximum filtering scale when the low-frequency features of the image are obvious

[0148] The specific steps of defect detection are:

[0149] Step 1: Gaussian filter construction:

[0150] 1. Construct a two-dimensional Gaussian low-pass filter, whose mathematical expression is:

[0151] in, is the frequency domain coordinate, is the filter radius.

[0152] 2. Discrete Fourier Transform:

[0153] For the input image Perform a 2D discrete Fourier transform to convert to the frequency domain:

[0154] in, are the number of rows and columns of the image, respectively.

[0155] 3. Frequency domain shift:

[0156] Move the frequency domain origin to the center of the image, the formula is:

[0157] 4. Frequency domain filtering:

[0158] Gaussian filter Multiplying the frequency domain data point by point after translation, the mathematical expression is:

[0159] 5. Inverse transformation and image reconstruction: Inverse frequency domain translation is the inverse operation of frequency domain translation, which restores the filtered frequency domain data to the original layout after discrete Fourier transform (low frequencies are on the periphery and high frequencies are in the center). Its mathematical formula is:

[0160] 6. Discrete Fourier Inverse Transform: The two-dimensional discrete Fourier inverse transform is to transform the filtered frequency domain data Convert back to spatial domain image The mathematical formula for this operation is:

[0161] 7. Calculate the amplitude: In order to obtain the final grayscale image that can be displayed, it is necessary to calculate the amplitude from the complex result containing the real part and the imaginary part. ( is the real part, is the imaginary part), its amplitude The calculation formula is:

[0162] 8. Normalization and type conversion: Normalize the image pixel values ​​to the range of 0-255. Mathematically, it can be simply understood as a linear transformation, which converts the original pixel value range to Map to The interval is as follows:

[0163] The original pixel value is , the normalized pixel value is .

[0164] Step 2: Threshold processing and mask creation:

[0165] A global thresholding operation is applied to the pre-processed image to generate a binary image based on the set threshold. Next, an erosion operation is performed to create a mask, using a structuring element of a specific shape to erode the binary image. This identifies the black areas in the binary image, which are used for region selection in subsequent operations.

[0166] Step 3: Adaptive threshold and morphological processing:

[0167] The image after preliminary processing is subjected to adaptive threshold processing, and a mean adaptive threshold algorithm is used to generate a binary image according to the set relevant parameters; then, the binary image is subjected to morphological operations of erosion and dilation to smooth the boundaries and remove noise.

[0168] Step 4: Connected component filtering:

[0169] The binary image is inverted and the connected component analysis function is used to detect the connected areas. A minimum area threshold is set to filter out connected components with an area smaller than the threshold and retain the valid areas.

[0170] Step 5: Edge detection and defect judgment:

[0171] Perform Canny edge detection on the processed image to generate an edge image; use the mask to perform a bitwise AND operation on the edge image to retain the edge information within the mask area; count the total number of white pixels in the edge image, and if it exceeds the set threshold, it is determined that there is a surface defect.

[0172] In addition, the specific steps for edge protrusion and depression detection are as follows:

[0173] Step 1: Threshold processing:

[0174] The input grayscale image is binarized based on the set first threshold. Through specific operations, the pixel values ​​in the image greater than the threshold are set to 255 (white), and the pixel values ​​less than or equal to the threshold are set to 0 (black). This preliminarily distinguishes the foreground and background of the image, laying the foundation for subsequent processing.

[0175] Step 2: Morphological dilation:

[0176] For the image after the above binarization processing, specific morphological operations are applied, and a structuring element with a size of 5×11 is used to expand the white area in the image to fill small holes, connect adjacent object parts, enhance the contour display of the target object, and reduce the inaccurate edge detection problem caused by the loss of image details in subsequent processing.

[0177] Step 3: Remove small connected components:

[0178] Perform a connected component analysis on the expanded image to obtain statistical information such as the number, label, area, and centroid coordinates of each connected component. A second dimension is set and all connected components except the background are traversed. Only connected components with an area larger than the second dimension are retained. Small noise points in the image are removed to obtain a relatively clean image, effectively improving the accuracy of subsequent edge detection.

[0179] Step 4: Edge detection:

[0180] For the image after multi-step preprocessing, the Canny algorithm is executed with the low threshold set to 100 and the high threshold set to 200 to accurately extract the edge information of the image and generate an edge image, in which the pixel value of the edge part is 255 and the pixel value of the non-edge part is 0.

[0181] Step 5: Image partitioning and feature analysis:

[0182] Image partitioning: Divide the edge image into two parts horizontally.

[0183] Grouping feature determination: Process the edge points in the left and right images separately. Extract the coordinates of the edge points and divide them into five roughly even groups. For each group of edge points, call the determination function (which determines whether the point meets the line fitting criteria) and pass in the edge point coordinates and tolerance parameters.

[0184] In addition, the specific steps of burr detection are:

[0185] Step 1: Gaussian low-pass filtering and threshold processing:

[0186] The same Gaussian filtering and frequency domain processing methods as those used for surface defect detection are used; a binary image mask is generated, and the region of interest is refined through two erosion operations.

[0187] Step 2: Edge detection and area restriction:

[0188] The cropped image is processed using the Canny edge detection algorithm to generate an image containing edge information. A "region of interest mask" is generated. This mask image is a binary image, where areas with pixel values ​​of 255 represent the image portion of interest, and areas with pixel values ​​of 0 represent areas of no interest. Edge detection is then performed on the original image to produce an "edge image." To focus the edge detection results on the region of interest, a bitwise AND operation is performed. This bitwise AND operation is a logical operation. For pixels at the same location in the two images, the corresponding pixel in the "edge image" is retained only if the pixel at that location in the "region of interest mask" is 255; otherwise, it is set to 0. This method eliminates edge information from areas of no interest, retaining only the edge information within the area specified by the "region of interest mask," ultimately resulting in a "restricted edge image."

[0189] Step 3: Morphological processing and connected component filtering:

[0190] To optimize subsequent image analysis, the "bounded edge image" requires further processing. First, a "dilation structuring element" with 5 rows and 5 columns is created. This structuring element determines the neighborhood range of the dilation operation and influences which pixels are considered neighbors. Next, a dilation operation is performed on the "bounded edge image." This connects edges that may have been disconnected due to noise or other factors, forming more continuous edge regions. The resulting "dilated edge image" is further processed. Furthermore, a minimum area threshold, designated "minimum area threshold," is set to 120 pixels. Next, a traversal operation is performed, which iterates over all connected regions except the background (background pixels are labeled 0 in the "connected region label matrix"). For each connected region, its area information recorded in the "connected region statistics array" is checked. If the area is greater than the "minimum area threshold," the connected region is marked white (with a pixel value of 255) in a new image (designated "cleaned image"). The purpose of this is to remove smaller connected regions, as these may be caused by noise or insignificant details in an image, while larger connected regions are more likely to be the actual glitches of interest. This screening method focuses attention on larger connected regions that are more likely to be glitches, allowing for further analysis or processing, thereby improving the accuracy and effectiveness of image detection and analysis.

[0191] Step 4: Result judgment:

[0192] The previously obtained "limited edge image" and "cleaned image" are added together to produce the final result image, named the "final result image." This step integrates the edge information from the "limited edge image" with the filtered connected region information from the "cleaned image," so that the final image contains both the original edge information and the filtered connected region information that may represent glitches, forming a comprehensive image result that facilitates subsequent judgment and analysis. Subsequently, the pixel sum of the "final result image" is calculated. If this pixel sum exceeds 1000, the image is considered defective, and the "defect flag" (defect_flag) is set to True. The threshold of 1000 is an empirical or experimental value determined through multiple experiments or based on actual conditions. When the pixel sum exceeds this threshold, it means that the "final result image" contains a high number of pixels that may represent glitches, indicating that the image may contain a high level of glitches. Therefore, it can be determined that the object corresponding to this image is defective and requires further attention or treatment.

[0193] Of course, the above are only typical examples of the present invention. In addition, the present invention may also have many other specific implementation methods. Any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

Claims

1. A ribbon defect detection system based on machine vision, characterized in that: Includes the following modules: The cropping module is used to remove the background from the image captured by the camera and divide the detection area. Specifically, it includes: Convert the color image into grayscale image and perform Gaussian filtering to remove noise; The ribbon outline is extracted by threshold segmentation and Canny edge detection, excluding the background area; The size detection module is used to detect the width, thickness and length of the ribbon and obtain the actual size data through image stitching and pixel ratio conversion; The character detection module is used to detect the fuzzy, offset, missing, incomplete, and deformed characters on the ribbon. Specifically, it includes: Grayscale conversion, Canny edge detection, connected region analysis and contour comparison; Density detection module, used to measure the average value of the warp and weft density of the ribbon and compare it with the standard value; Defect detection module, including surface defect detection, edge protrusion and depression detection, and burr detection, specifically using Gaussian low-pass filtering, threshold processing, morphological operations and connected component analysis; The data analysis module group includes the size data analysis module, character data analysis module, density data analysis module and defect data analysis module, which input the inspection data into the corresponding model to generate the size coefficient , character coefficient , density coefficient and defect coefficient ; , They are length deviation, width deviation and thickness deviation, are the weights of length, width, and thickness deviations, respectively; , is the measured density, is the standard density; , The coefficients of influence of character fuzziness, character position offset, missing or incomplete characters, and character deformation on the character quality coefficient are respectively: Indicates that the characters on the ribbon are blurred, the characters are offset, the characters are missing or incomplete, or the characters are deformed; , These are the influence coefficients of surface defects, edge protrusions and depressions, and burrs on the overall defect coefficient of the ribbon. Indicates surface defects, Indicates edge protrusions and depressions, Indicates burr defects; Comprehensive data analysis module, used to integrate various coefficients to generate ribbon quality evaluation index And compare it with the standard value to determine whether the webbing is qualified; The result judgment and information interaction module is used to display the test results and trigger early warning signals based on the quality assessment index.

2. A machine vision-based ribbon defect detection system according to claim 1, characterized in that: The specific steps of the size detection module include: The edge detection results are used to extract the ribbon contour and the width is calculated by the horizontal boundary points. ; Calculate the average thickness by the vertical distance between the upper and lower edge points ; Measure the total length of the ribbon by stitching multiple images through SURF feature point matching and affine transformation .

3. The machine vision-based ribbon defect detection system according to claim 1, characterized in that: The defect detection module further includes an adaptive adjustment mechanism that can dynamically optimize detection thresholds and light source parameters based on the ribbon material, color and historical data.

4. The machine vision-based ribbon defect detection system according to claim 1, characterized in that: The surface defect detection comprises the following steps: Filter the image texture in the frequency domain using a Gaussian low-pass filter; Global thresholding, morphological operations, and connected component filtering; The pixel sum is calculated through edge detection and mask operation to determine the presence of defects.

5. The machine vision-based ribbon defect detection system according to claim 1, characterized in that: The edge protrusion and depression detection comprises the following steps: Binarize and morphologically dilate the grayscale image; Perform Canny edge detection after removing small connected components; The edge images are partitioned and grouped for straight line fitting analysis.

6. The machine vision-based ribbon defect detection system according to claim 1, characterized in that: The burr detection comprises the following steps: Generate a mask and define the edge detection area; connecting fracture edges through dilation operations; Small area connected components are filtered and burr defects are determined by pixel sum threshold.

7. The machine vision-based ribbon defect detection system according to claim 1, characterized in that: The quality assessment index of the comprehensive data analysis module The judgment rules are: when The webbing is judged to be qualified; when When the warning signal is triggered; in is the standard quality index, is the allowable deviation ratio.

8. A method for detecting ribbon defects based on machine vision according to any one of claims 1 to 7, characterized in that: The following steps are involved: Divide the detection area through the cropping module; Perform size, character, density and defect inspections sequentially and collect data; Input the data into the corresponding analysis model to generate coefficients; Comprehensively evaluate the quality of the webbing by combining various coefficients and output the results.

Images