Braided tube appearance defect detection method and system based on machine vision

Abstract

The invention relates to the technical field of defect identification, in particular to a braided tube appearance defect detection method and system based on machine vision, and the method comprises the following steps: obtaining an image to generate a brightness fracture region, calculating a peak difference to generate a structure abnormal region, extracting a light spot path to identify a burr abnormal region, and analyzing a grid texture to generate a deformation identification region. And a defect identification scheme is obtained by combining gray scale and jump. According to the method, a brightness fracture area is extracted through gray difference symbol change, segmentation of image jump features is realized, a structure dislocation area is positioned through symmetrical wave crest difference, a form mutation zone is revealed through light spot extension and edge consistency comparison, and burr identification stability is improved. Local trend abnormity is judged through texture length and cross frequency quantification, and a multi-dimensional standard is constructed by combining gray span and disturbance frequency screening, so that defect features are clearly presented in a detail level, and the recognition precision and stability in a complex surface environment are ensured.

Description

Technical Field

[0001] This invention relates to the field of defect recognition technology, and in particular to a method and system for detecting appearance defects in braided tubes based on machine vision. Background Technology

[0002] Defect recognition technology is an interdisciplinary application of image processing and pattern recognition. It primarily encompasses the detection, classification, and identification of various visual defects such as cracks, dents, scratches, foreign objects, and stains on the surfaces of products, materials, or components using machine vision, image acquisition and processing, and computer recognition algorithms. It is a crucial link in achieving automated quality control in industry and is widely used in sectors such as electronics manufacturing, metal processing, textiles, and pharmaceutical packaging, playing a key role, especially in achieving efficient, objective, and continuous quality monitoring. Its core aspects include the configuration and calibration of image acquisition equipment, image preprocessing and feature extraction, defect sample training, and identification and classification methods. Pattern recognition and pattern matching technologies form the technological foundation for defect recognition. Traditional methods for detecting defects in braided tubing refer to the methods used to detect defects in the appearance of metal or polymer fiber braided tubing used for fluid transmission. These defects are caused by uneven materials, abnormal tension, or errors in the braiding process. The methods are usually completed by manual visual inspection. The main basis is that the operator identifies defects by visual inspection based on experience and standards, and relies on manual recording and sampling for quality assessment. This method has problems such as low efficiency, poor stability, and insufficient repeatability.

[0003] The shortcomings of existing technologies are that the precision of defect identification by manual visual inspection depends on visual discrimination ability. Under complex image textures or multi-layered interweaving conditions, it is easy to make detailed misjudgments. The identification of defect boundaries relies on experience judgment, which leads to ambiguity in the definition of fracture and deformation areas. When the contrast of peaks and the changes in texture direction are subtle, it is difficult for manual methods to establish stable quantitative standards. In long-term inspection environments, continuous defects are easily missed. The recording method is mainly based on sampling inspection, which leads to overall data loss. As a result, it is difficult to track the clustering characteristics and distribution patterns of complex defects, thus causing monitoring lag and quality feedback delays in production scenarios. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a machine vision-based method for detecting appearance defects in braided tubing, comprising the following steps: S1: Acquire blue light interference image frames, extract scan lines along the radial direction of the image center, locate the gray peaks of the scan lines, construct the brightness change trend through the gray difference of the peaks, extract the sign jump breakpoints, and generate brightness break recognition areas. S2: Select the symmetrical direction scan line in the brightness fracture identification area, calculate the corresponding transverse difference value and dense distribution of the peak, locate the difference aggregation area, map it to the radial yarn interface structure of the braided tube, and generate the structural abnormal area. S3: Analyze the structurally abnormal region, extract the light spot abrupt change zone and the peak extension path, measure the expansion trend on both sides of the fitted line, screen regions that are inconsistent with the edge direction, identify the phenomenon of continuous morphological jump, and generate burr abnormality identification region. S4: Analyze the image grid of the burr anomaly recognition area, extract the gray-scale direction change sequence and perform linear segmentation, count the duration and cross frequency, and generate the deformation recognition area by referring to the diagonal stripe direction as a reference. S5: Extract the deformation recognition area, measure the grayscale span and disturbance frequency, analyze the area where the texture and brightness change are consistent, and combine the boundary grayscale break and continuous jump to obtain the braided tube appearance defect recognition scheme.

[0005] As a further aspect of the present invention, the brightness breakage identification region includes breakpoint distribution density, grayscale difference sequence, and brightness change trend; the structural anomaly region includes lateral difference value distribution, offset difference aggregation, and misaligned structural bands; the burr anomaly identification region includes spot abrupt change bands, peak extension characteristics, and morphological jump phenomena; the deformation identification region includes grayscale direction continuity length, texture crossover frequency, and directional anomaly segments; and the braided tube appearance defect identification scheme includes texture direction consistent segments, boundary grayscale breakage characteristics, and continuity jump degree.

[0006] As a further aspect of the present invention, the grayscale peak of the scan line refers to the pixel position where the grayscale value reaches and is higher than that of the adjacent point along the scan line direction.

[0007] As a further aspect of the present invention, the difference aggregation region refers to the area where the lateral difference values ​​of corresponding peaks on adjacent scan lines are densely distributed and concentrated in space.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire a blue light interference image frame, extract scan lines divided along the radial direction of the image center, call the gray level in the image frame, locate the local gray maximum value on each scan line, and generate a gray peak localization sequence. S102: Call the grayscale peak positioning sequence, calculate the grayscale difference between adjacent peaks, construct the symbol change sequence, extract the symbol jump position, and generate a jump breakpoint segment index group. S103: Based on the jump breakpoint segment index group, count the number of breakpoints within a unit pixel, determine the breakpoint density distribution, filter out areas where the density exceeds the threshold, and obtain the brightness break recognition area.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the scan line located in the symmetrical direction of the image in the brightness fracture recognition area, retrieve the gray peaks with the same number on the left and right scan lines, obtain the coordinate values ​​of the symmetrical peaks on the horizontal axis, calculate the horizontal coordinate difference, and arrange all the differences in order to generate a horizontal difference value sequence. S202: Based on the horizontal difference value sequence, divide the data into segments according to the preset coordinate difference distribution interval, count the number of differences in the interval, determine whether the distribution density exceeds the horizontal difference cluster density threshold, record the corresponding interval index position, and obtain the offset difference cluster index group. S203: Based on the mapping relationship between the offset difference aggregation index group and the image frame to the braided tube structure, retrieve the corresponding radial yarn interface area, mark the area range where there is symmetrical offset, extract the structural boundary information of the area, and obtain the structural abnormal area.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Call the structural anomaly region, extract the gray-level change trajectory in the boundary direction, detect the position of gray-level abrupt change points in the trajectory, extract the brightness enhancement region formed by the continuous arrangement of gray-level abrupt change points, obtain the light spot structure distributed along the edge direction, and establish a light spot abrupt change band path set. S302: Based on the set of light spot abrupt change zone paths, fit the extension direction straight line of each path, and calculate the extension width difference index of the peak distribution on both sides of the path. Make a difference judgment on the left and right extension widths, and filter out the paths whose extension offset direction is not asymmetrical with the edge direction to obtain the region group with inconsistent extension direction. S303: Call the boundary continuous point list of each region in the inconsistent extension direction region group, detect the gray level jump amplitude and jump number of adjacent pixels point by point, determine whether there is continuous fluctuation across the jump amplitude threshold in the gray level response sequence, mark the boundary segment of the region that meets the jump feature, and obtain the spur anomaly identification region.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the burr anomaly recognition area, obtain the pixel grayscale values ​​along the row and column directions in each image grid, construct the grayscale change sequence in the horizontal and vertical directions, perform direction extraction operation on the continuous gradient change segments in each group of sequences, divide according to the direction change interval, and obtain the local texture direction sequence set; S402: Based on the local texture direction sequence, the duration of each direction change and the number of times the intersection direction appears are counted to form the direction change feature vector of each grid. The direction information of the diagonal cross area on the surface of the braided tube is used as a reference to calculate the direction difference value between the feature vector and the reference direction and obtain the texture direction difference matrix. S403: Based on the distribution of orientation differences in the grids in the texture orientation difference matrix, determine whether the orientation change exceeds the orientation difference threshold in the continuous grids, mark the continuous segments of abnormal changes, extract the index of the covered image region, and obtain the deformation recognition region.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the edge image of the patch in the deformation recognition area, extract the set of boundary pixels of each patch, detect the gray value of the boundary pixels, calculate the gray range span index, and construct a gray gradient vector field in combination with the distribution direction of the boundary pixels. Measure the directional perturbation frequency of the vector field and obtain the gray direction perturbation sequence. S502: Based on the gray-scale direction perturbation sequence, calculate the difference between the perturbation change frequency and the boundary extension direction, perform interval statistics on the amplitude sequence, filter segments with amplitude within the direction difference threshold, extract the local texture direction information of the segments, and determine whether the texture and brightness change direction are consistent to obtain a set of texture segments with consistent direction. S503: Based on the set of texture fragments with consistent direction, call the continuous jump amplitude and boundary gray-scale breakage feature data in the fragments, perform joint judgment on the two, filter out the regions where the continuous jump amplitude exceeds the jump baseline threshold and is accompanied by gray-scale breakage, aggregate the spatial index range of the regions, and obtain the braided tube appearance defect identification scheme.

[0013] A machine vision-based braided tubing appearance defect detection system includes: The brightness break detection module is used to achieve S1: acquiring blue light interference image frames, extracting scan lines along the radial direction of the image center, locating the gray peaks of the scan lines, constructing the brightness change trend through the gray difference of the peaks, extracting the sign jump breakpoints, and generating the brightness break recognition area. The structural anomaly localization module is used to implement S2: select the symmetrical direction scan line in the brightness fracture identification area, calculate the corresponding transverse difference value and dense distribution of the peak, locate the difference accumulation area, map it to the radial yarn interface structure of the braided tube, and generate the structural anomaly area; The burr anomaly detection module is used to achieve S3: analyze the structural anomaly region, extract the light spot abrupt change zone and peak extension path, measure the expansion trend on both sides of the fitted line, screen regions that are inconsistent with the edge direction, identify continuous morphological jump phenomena, and generate burr anomaly recognition region. The deformation trend recognition module is used to implement S4: analyze the image grid of the burr anomaly recognition area, extract the gray-scale direction change sequence and perform linear segmentation, count the duration and cross frequency, and generate the deformation recognition area by referring to the diagonal stripe direction as a reference. The defect comprehensive identification module is used to implement S5: extract the deformation identification area, measure the gray scale span and disturbance frequency, analyze the area with consistent texture and brightness changes, and combine the boundary gray scale breakage and continuous jump to obtain the braided tube appearance defect identification scheme.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, brightness fracture regions are extracted by changing the sign of grayscale difference, thus segmenting image jump features. Structural misalignment regions are located by symmetrical peak differences. Morphological abrupt change zones are revealed by comparing the extension of light spots with edge consistency, improving the stability of burr recognition. Local directional anomalies are determined by quantifying texture length and cross frequency. Multi-dimensional standards are constructed by combining grayscale span and perturbation frequency screening, so that defect features are clearly presented at the detail level, ensuring recognition accuracy and stability in complex surface environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a machine vision-based method for detecting appearance defects in braided tubing, comprising the following steps: S1: Acquire blue light interference image frames, extract scan lines divided along the radial direction of the image center, call the gray level in the image frame, locate the local brightness peaks of the scan lines, construct a brightness change trend sequence through the gray difference sequence of adjacent peaks, extract the jump breakpoint segments based on the sign change direction, determine the density distribution of continuous breakpoints, and extract the brightness break recognition area. S2: Call the scan line located in the symmetrical direction of the image in the brightness fracture recognition area, compare the position coordinates of the same number peak in the left and right scan lines, calculate the horizontal difference value sequence, and count the density of the distribution interval. Locate the large-scale cluster area of ​​offset difference, screen the corresponding area in the radial yarn interface structure of the braided tube through the mapping relationship, identify the misaligned structure band, and generate the structural abnormal area. S3: Call the boundary image region included in the structural anomaly region, extract the light spot abrupt change band in the boundary direction, obtain the brightness peak extension path along the edge direction, measure the expansion trend of the peak response on both sides of the fitting line, filter the region where the extension offset direction is inconsistent with the edge direction, identify the continuous jump phenomenon in the morphological stability of the region, and generate the spur anomaly identification region. S4: Call the burr anomaly identification area, obtain the pixel gray-level direction change sequence in the grid, perform linear segmentation processing on the local texture trend, count the gray-level direction continuity length and cross frequency, call the direction of the diagonal cross area on the surface of the braided tube as the stability reference benchmark, judge the abnormal segment of the continuity of the image grid texture direction change, and generate the deformation identification area. S5: Call the image of the patch edge in the deformation recognition area, obtain the grayscale range of the patch, and measure the difference between the frequency of directional disturbance change and the boundary extension shape. Based on the trend of image edge directional structural change, filter local segments with the same direction of texture and brightness change. Combine the degree of continuous jump and the boundary grayscale breakage feature to obtain the braided tube appearance defect recognition scheme.

[0023] The brightness breakage identification area includes the breakpoint distribution density, grayscale difference sequence, and brightness change trend; the structural anomaly area includes the lateral difference value distribution, offset difference aggregation, and misaligned structural bands; the burr anomaly identification area includes the spot abrupt change band, peak extension characteristics, and morphological jump phenomenon; the deformation identification area includes the grayscale direction continuity length, texture cross frequency, and directional anomaly segments; and the braided tube appearance defect identification scheme includes texture direction consistent segments, boundary grayscale breakage characteristics, and continuity jump degree.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire a blue light interference image frame, extract scan lines divided along the radial direction of the image center, call the gray level in the image frame, locate the local gray maximum value on each scan line, and generate a gray peak localization sequence. First, optical acquisition equipment is used to acquire continuous interference pattern image frames on the surface of the object being measured. These image frames typically have different resolutions or sizes after initial acquisition, requiring image standardization to a size of, for example, 512×512 pixels. This ensures that all subsequent image processing is performed under the same spatial reference. Then, starting from the geometric center of the image, multiple scan lines are radially divided towards the image edges. The number of scan lines is determined by the angular interval between each line; for example, one scan line is divided every 10 degrees, resulting in 36 scan lines for the entire image. These scan lines pass through the center and extend in a specified direction. On each scan line, the grayscale value is read point-by-point from the image center outwards, unit by unit, to construct... The grayscale sequence corresponding to the scan line contains integers between 0 and 255 for each grayscale value. Then, by comparing the grayscale values ​​of three consecutive pixels on each grayscale sequence, it is determined whether the intermediate point is a local maximum. If the grayscale value of a point is greater than the grayscale values ​​of its two adjacent points, it is identified as a local grayscale maximum and its position and corresponding grayscale value on the scan line are recorded. For example, if three consecutive grayscale values ​​on a scan line are 65, 90, and 75, then the position 90 is identified as a local grayscale maximum. This process is repeated in all scan lines to ensure that the local grayscale maximums on each line are accurately extracted. Finally, the set of maximum values ​​extracted from each scan line is summarized to form a complete grayscale peak localization sequence of the image frame.

[0025] S102: Call the grayscale peak positioning sequence, calculate the grayscale difference between adjacent peaks, construct the symbol change sequence, extract the symbol jump position, and generate the jump breakpoint segment index group; For each scan line, take any two adjacent peaks in the sequence and calculate their grayscale difference. After extracting the grayscale values ​​of the two peaks, obtain the difference for that segment through a simple subtraction operation. For example, if two consecutive peaks on a scan line have grayscale values ​​of 110 and 125, then the grayscale difference for that segment is 15. Calculate and record the difference for each pair of adjacent peaks to form a difference list. In the difference list, determine whether each difference is positive, negative, or zero. If it is greater than zero, the sign at that position is positive; if it is less than zero, it is negative; and if it is zero, it is marked as neutral. Continue to label these signs to form a complete sign change sequence. This sign sequence reflects the direction of the grayscale value change trend. Next, by traversing the symbol change sequence, it identifies whether there is a change between two adjacent symbols, that is, a jump where the previous symbol is positive and the next is negative or vice versa. The peak index corresponding to the position where the symbol changes is recorded as the jump breakpoint. For example, if the symbol sequence is positive, positive, negative, positive, then the symbol jumps from positive to negative in the second to third item and from negative to positive in the third to fourth item. The peaks at these positions are recorded as jump breakpoints. The above operation is performed on all scan lines in the image. Finally, the set of breakpoint positions corresponding to the jump symbols is extracted. All the identified jump breakpoints are summarized to generate a jump breakpoint segment index group.

[0026] S103: Based on the jump breakpoint segment index group, count the number of breakpoints within a unit pixel, determine the breakpoint density distribution, filter out areas with density exceeding the threshold, and obtain the brightness breakpoint recognition area. First, the entire image frame needs to be divided into fixed regions, such as using an 8×8 pixel sliding window. The entire image region is traversed sequentially, with the sliding window moving from left to right and from top to bottom in 8-pixel increments. For each window, the number of transition breakpoints within that window area is counted. This is determined by reading the spatial coordinates of the breakpoints and checking if they fall within the pixel boundaries of the current window. If so, the breakpoint count for that window is incremented. After counting breakpoints for all windows in the entire image, a breakpoint density distribution map corresponding to the image size is obtained. The breakpoint density for each window is the number of breakpoints divided by the window area. For example, if 6 transition breakpoints are recorded within an 8×8 pixel window, the breakpoint density for that window is 6 / 64, approximately 0.09375. Next, the breakpoint densities of all windows are statistically analyzed, calculating the average and standard deviation of the breakpoint densities for all windows in the image. For example, the average density is 0.05, and the standard deviation is 0.015. Based on this, a threshold ρ is set. t Add twice the standard deviation to the average density, i.e., ρ t=0.05+2×0.015=0.08, this threshold represents the identification benchmark for dense breakpoint regions. Then, the breakpoint density value of each window is traversed. If its density is higher than 0.08, the window is marked as a break region. All window regions that meet the density higher than the threshold are merged to form a complete brightness break recognition region.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the scan line located in the symmetrical direction of the image in the brightness fracture recognition area, retrieve the gray peaks with the same number on the left and right scan lines, obtain the coordinate values ​​of the symmetrical peaks on the horizontal axis, calculate the horizontal coordinate difference, and arrange all the differences in order to generate a horizontal difference value sequence. First, determine the symmetry reference direction of the image, usually the vertical direction from the geometric center of the image as the axis of symmetry. By reading the image width value, calculate its center column coordinates. Pair the scan lines on both sides of the center line with the same angle or row spacing, limiting the matching range to the brightness break detection area. Then, within each pair of symmetrical scan lines, sequentially search for peaks with the same index according to the peak sequence numbering. Read the lateral coordinate position of each peak and store the corresponding peak coordinates on the left and right scan lines. For example, if the lateral coordinate of the 3rd peak on the left scan line is 120 pixels and the lateral coordinate of the 3rd peak on the right scan line is 140 pixels, the corresponding lateral difference is 20 pixels. Continue in this manner, calculating the lateral coordinate difference between each pair of peaks, ensuring that the order of all differences matches the peak indexing order. Peaks that cannot find a symmetrical match are discarded to ensure the complete pairing of the difference sequence. If 10 valid peak pairs are detected in a set of symmetrical scan lines, 10 horizontal difference data will be obtained. These differences are arranged sequentially from the center outwards. For example, a set of horizontal difference values ​​such as 12, 15, 18, and 20 can be obtained to form a complete horizontal difference value sequence.

[0028] S202: Based on the horizontal difference value sequence, the difference value is divided into segments according to the preset coordinate difference value distribution interval, the number of differences in the interval is counted, and it is determined whether the distribution density exceeds the horizontal difference value clustering density threshold. The corresponding interval index position is recorded, and the offset difference value clustering index group is obtained. First, determine the possible range of horizontal differences. For example, based on the image size, set the maximum horizontal difference to 50 pixels. Then, divide this range into sections with a fixed width, such as 5 pixels per section, resulting in 10 distribution sections. Iterate through the sequence of horizontal difference values, assigning each difference to its corresponding section based on its numerical range. For example, a difference of 8 pixels is recorded in the [6, 10] section, and a difference of 17 pixels is recorded in the [16, 20] section, and so on, completing the distribution classification of the entire sequence. Next, count the number of differences within each section to determine the degree of clustering for that section. To determine which sections have a significant concentration of differences, a horizontal difference clustering density threshold needs to be set. The threshold is set by averaging the number of differences across all sections, and then adding a compensation constant based on the image structure variation characteristics.

[0029] The compensation constant is typically set to 30% to 50% of the average value. For example, when each interval contains an average of 3 differences, the compensation constant can be set to 1.5, resulting in a threshold of 4.5, which is rounded to 5. Each interval is iterated over; when the number of differences in an interval exceeds 5, it is considered a dense cluster of differences, and the interval's index number is recorded. For example, if 7 differences are found in the interval [11, 15], this interval is marked as a clustered offset difference interval. All interval numbers that meet the criteria are arranged sequentially to form an offset difference cluster index group, which is used for subsequent structural offset position analysis.

[0030] S203: Based on the mapping relationship between the offset difference cluster index group and the image frame to the braided tube structure, retrieve the corresponding radial yarn interface area, mark the area range where there is symmetrical offset in the marked position, extract the structural boundary information of the area, and obtain the structural abnormal area. First, based on the mapping table established before image acquisition, the correspondence between the horizontal axis coordinates of the image and the actual spatial positions of the radial yarns in the braided tube structure is determined. Each scan line number corresponds to the interface position of a specific radial yarn. The interval numbers recorded in the offset difference aggregation index group are read to determine the scan line number and position coordinates of the peaks where the horizontal differences correspond. The mapping table is searched item by item to find the corresponding radial yarn regions in the actual braided structure for these horizontal difference aggregation points. For example, if the peak position corresponding to aggregation area number 4 in the index group is located at image coordinates (x=145, y=220), the table shows that this position corresponds to the interlacing interface region of the 10th radial yarn. Then, with this position as the center in the image, the grayscale changes of adjacent pixels are read along the radial and horizontal directions. Based on the continuity of the grayscale changes, the boundary range of this region is determined. Border pixels with significant grayscale changes are connected to form a closed region to define the structural outline. The above steps are repeated for all marked regions to finally obtain the boundary coordinate ranges of all radial regions with symmetrical offsets, thereby extracting the specific regions of structural anomalies.

[0031] Please see Figure 4 The specific steps of S3 are as follows: S301: Call the structural anomaly region, extract the gray-level change trajectory in the boundary direction, detect the position of gray-level abrupt change points in the trajectory, extract the brightness enhancement region formed by the continuous arrangement of gray-level abrupt change points, obtain the light spot structure distributed along the edge direction, and establish a light spot abrupt change zone path set. First, it is necessary to extract a continuous series of points along the boundary of the region and determine the direction of the boundary. For each boundary line segment, extract the gray-level change trajectory along its direction. Specifically, select sampling points with a one-pixel interval on the boundary line, and sequentially read the gray-level values ​​of multiple adjacent pixels along the normal direction of each sampling point, forming a gray-level change curve along the boundary direction. Scan the gray-level changes of these gray-level trajectories sequentially, comparing the gray-level difference of adjacent pixels point by point. If the change in gray-level value at a certain point is significantly greater than the average gray-level change level compared to the previous point, then that point is determined to be a gray-level abrupt change point. The gray-level abrupt change threshold is set based on calculating the average gray-level change amplitude of the overall gray-level distribution of the structurally abnormal region, and then setting an adjustment amount according to the brightness range and noise level of the image. For example, when the average gray-level change is 8 gray levels, and the image noise is controlled within ±2 gray levels, the threshold can be set to 10. When multiple consecutive gray-level abrupt change points are arranged along the boundary direction and the distance between them does not exceed two pixels, these points are merged and regarded as a brightness enhancement region. Following this method, the boundaries of the entire structurally abnormal region are processed one by one to obtain multiple continuous paths of brightness enhancement regions. Finally, all paths are numbered and recorded to form a light spot structure distributed along the edge direction, and a light spot abrupt change zone path set is generated on this basis.

[0032] S302: Based on the path set of light spot abrupt change zone, fit the straight line of the extension direction of each path, and calculate the extension width difference index of the peak distribution on both sides of the path. Make a difference judgment on the left and right extension widths, and screen out the paths whose extension offset direction is not asymmetrical with the edge direction to obtain the region group with inconsistent extension direction. The specific formula for calculating the difference in the extension width of the wave crest distribution on both sides of the path is as follows: ; Calculate the extension width difference index, which is used to determine whether there is a significant difference between the left and right extension widths. The difference between the left and right extension widths is judged, and paths whose extension offset direction is not asymmetrical with the edge direction are filtered to obtain a group of regions with inconsistent extension directions. in, An index representing the difference in the extension width of the wave crest distribution on both sides of the path. Indicates the leftmost path The vertical distance from each peak point to the fitted line. This represents the average vertical distance from the left-hand peak. Indicates the rightmost path The vertical distance from each peak point to the fitted line. This represents the average vertical distance from the right-hand peak. Indicates the number of peaks on the left side. Indicates the number of peaks on the right side. A constant representing a very small positive number, used to prevent the denominator from being zero; The numerator of the formula reflects the difference in variance of the distribution of the left and right peaks through the difference of the sum of the two squared differences, that is: Left side: ; Right side: ; The difference obtained by subtracting the two is expressed in absolute value to ensure a positive result. The denominator uses... Normalize the molecules to reduce interference between different path scales. (Proportional term) This is used to balance the effects of the asymmetry in the number of wave peaks on both sides, and to enhance the formula's sensitivity to structural changes.

[0033] The actual process of obtaining and assigning parameters is as follows: A light spot path is selected in the image, and an automatic grayscale peak extraction algorithm identifies 5 peaks on the left and 4 peaks on the right. The distance from each point to the fitted line is measured in pixels. Vertical distances between left-side peaks (in pixels): 12, 15, 14, 13, 16; Calculate the average value on the left: ; Vertical distances between right-side peaks (in pixels): 10, 9, 11, 13; Calculate the average value on the right: ; Number of left-side peaks: ; Number of peaks on the right: ; Minimal constant: This value sets the minimum unit interval for the reference image resolution, taking the minimum effective pixel unit to prevent division by zero; Substitute the above parameters into the formula to perform the calculation: Sum of left-side squared deviations: ; Sum of right-hand squared deviations: ; Molecular difference: ; Normalized term for denominator: ; Quantity balance item: ; The final calculation result is: ; The results show that the difference in peak extension between the left and right sides of the path is 0.3195 pixels. If the system sets the extension consistency tolerance threshold to 0.3 pixels, then this path exceeds the tolerance threshold and is determined to be a "path with inconsistent extension direction". It will be included in the inconsistent extension direction region group for subsequent burr abnormal structure aggregation analysis.

[0034] The advantage of the formula is that, by introducing a quantity ratio correction factor and a variance comparison normalization mechanism, it can achieve stable judgment of symmetrical structure under conditions of unequal number of peaks or uneven distribution density, improve the reliability of asymmetric feature recognition, and enhance the system's ability to distinguish structural offsets in the edge direction.

[0035] Table 1. Statistics of Vertical Distance Between Gray Scale Peaks Serial Number Distance between left-side peaks (pixels) Distance between right-side peaks (pixels) 1 12 10 2 15 9 3 14 11 4 13 13 5 16 —— Table 1 lists the vertical distance data between the grayscale peaks on the left and right sides of the implementation path and the straight line of the fitted extension direction, providing the basic data support required for index calculation. Combined with the above formula calculation and comparison, the path direction consistency screening and judgment were completed.

[0036] S303: Call the continuous point list of the boundary of each region in the inconsistent extension direction region group, detect the gray level jump amplitude and jump number of adjacent pixels point by point, determine whether there is a continuous fluctuation that crosses the jump amplitude threshold in the gray level response sequence, mark the boundary segment of the region that meets the jump feature, and obtain the spur anomaly identification region. Read the continuous boundary point series of each region and perform point-by-point grayscale change detection on these point series. Compare each boundary point with its neighboring pixels, record the grayscale difference, and generate a grayscale jump sequence in sequence. Then, count the jump amplitude and number of jumps for each boundary segment. If the grayscale difference exceeds the set jump amplitude threshold, it is recorded as a valid jump. The jump amplitude threshold is set based on the average amplitude of grayscale change across the entire image and the local noise level. When the average grayscale change amplitude across the entire image is 6 gray levels and the local noise is approximately 3 gray levels, the threshold can be set to 9. Next, check the continuous jump situation on each boundary segment. When the number of continuous jump points exceeds three and the interval between adjacent jump points is less than three pixels, it is determined that there is obvious grayscale fluctuation in the boundary segment. Further check whether the grayscale change direction of these jump points alternates, that is, from increase to decrease and then from decrease to increase. If such alternation exists and the amplitude continuously exceeds the threshold, it is recorded as a boundary segment with spur characteristics. Number all boundary segments that meet this condition and extract their coordinate ranges. Summarize these to form a burr anomaly identification area, which is used to identify the location distribution of local structural detail anomalies.

[0037] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the spur anomaly identification region, obtain the pixel grayscale values ​​along the row and column directions in each image grid, construct the grayscale change sequence in the horizontal and vertical directions, perform direction extraction operation on the continuous gradient change segments in each group of sequences, divide according to the direction change interval, and obtain the local texture direction sequence set; First, based on the coordinate range of the abnormal region in the image, the image is divided into grid units with fixed pixels, for example, each grid is set to 16×16 pixels. The grayscale information of all pixels within each grid is extracted one by one. Then, the grayscale values ​​of each row and column of pixels are sampled according to the row and column directions, respectively, forming horizontal and vertical grayscale sequences. For each grayscale sequence, the grayscale difference between two adjacent pixels is compared sequentially from left to right or from top to bottom, recording the grayscale change trend, i.e., the direction information of grayscale value increase, decrease, or remain unchanged. Segments with the same continuous change trend are tracked to identify continuous gradient change segments where grayscale continuously increases or decreases. Each continuous grayscale change segment is defined as a direction segment. The relative direction between the start and end points of the segment is taken as the grayscale direction of the segment. Then, the angle formed by the direction is divided into fixed direction intervals. For example, 0°~180° is divided into six 30-degree intervals, 0°~30° is the first direction interval, 31°~60° is the second direction interval, and so on. Multiple directional segments extracted from each grid are assigned to their respective directional intervals, and these directional segments are organized to form a local texture direction sequence for that grid. By repeating this process for multiple grids, the respective direction sequences are summarized, ultimately forming a set of local texture direction sequences covering the entire burr anomaly region.

[0038] S402: Based on the local texture direction sequence, the duration of each direction change and the number of times the intersection direction appears are counted to form the direction change feature vector of each grid. The direction information of the diagonal cross area on the surface of the braided tube is used as a reference to calculate the direction difference value between the feature vector and the reference direction and obtain the texture direction difference matrix. The orientation sequence of each image grid is processed one by one. For an orientation sequence in a grid, according to the change of orientation angle, segments with continuous orientation changes and stable angles are identified. Multiple orientation segments that remain continuously within the same orientation interval are merged into a linear segment, and the start and end positions and orientation angle interval of each segment are recorded. Then, the number of pixels that maintain the orientation is calculated for each linear segment, the duration of each segment is counted, and the number of times the orientation changes from the main orientation to its perpendicular or diagonal orientation within each segment is detected. The orientation maintenance length of each segment and the number of cross-direction jumps are combined to form an orientation change feature vector. The first dimension of the feature vector is the main orientation maintenance length, and the second dimension is the number of times the cross-direction appears. Then, the orientation information of the diagonal cross-region on the surface of the braided tube is used as a comparison reference. This information is obtained according to the braided structure design and is usually a 45-degree or 135-degree diagonal orientation. The angle difference between the main orientation and the reference orientation of each grid is compared. The angle difference is the orientation difference value. When the angle difference is less than 10 degrees, it is judged as consistent; 10-25 degrees is a deviation; and more than 25 degrees is a serious deviation. The orientation differences of each grid are recorded as corresponding matrix elements. Finally, the orientation differences of all grids are organized into a texture orientation difference matrix for subsequent regional trend judgment.

[0039] S403: Based on the distribution of orientation differences in the grid in the texture orientation difference matrix, determine whether the orientation change exceeds the orientation difference threshold in the continuous grid, mark the continuous segments of abnormal changes, extract the index of the covered image region, and obtain the deformation recognition region. Based on the direction difference value recorded in each grid of the texture direction difference matrix, all grid elements are traversed, and grids with direction differences exceeding a set threshold are marked. The direction difference threshold is determined according to the allowable texture variation range of the woven structure. In a diagonal weave structure, an angle error of 10°~15° is usually allowed. To more accurately detect abnormal direction changes, the threshold is set to 20°. That is, if the direction difference of a grid is greater than 20°, it is considered that a direction abnormality has occurred at that point. Subsequently, a sliding window scan is performed on the entire difference matrix. The window size can be set to 3×3 or 5×5 to detect whether multiple adjacent grids continuously exceed the direction difference threshold. When the direction difference of three or more consecutive grids is found to exceed the set threshold in any direction, the group of grids is marked as an abnormal variation segment. The marked segments are further examined for their coverage area in the image. The corresponding image region is extracted according to the mapping relationship between the grid position and the actual pixel coordinates of the image. These continuous coverage areas are integrated into one or more deformation recognition blocks, and their grid number, image coordinate range, direction abnormality trend, and other information are recorded to complete the identification of abnormal continuous texture deformation regions.

[0040] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the edge image of the patch in the deformation recognition area, extract the set of boundary pixels of each patch, detect the gray value of the boundary pixels, calculate the gray range span index, and construct a gray gradient vector field in combination with the distribution direction of the boundary pixels. Then, measure the directional perturbation frequency of the vector field and obtain the gray direction perturbation sequence. First, boundary extraction is performed on each identified patch. A pixel-by-pixel scanning method is used to identify locations where grayscale changes occur between the patch and adjacent areas. Then, the set of boundary pixels constituting the patch outline is extracted. After extraction, the grayscale value of each pixel in this set is read, and the maximum and minimum grayscale values ​​are recorded. The difference between these two values ​​is calculated as the grayscale range index. If this range exceeds 1.5 times the average grayscale fluctuation value of the entire image, the boundary grayscale is considered to have a large degree of change. For example, if the average grayscale fluctuation of the entire image is 12, and the range judgment reference value is set to 18, if the grayscale range of the patch boundary is 20, it is considered... It exhibits abrupt changes in boundary features. Subsequently, taking each boundary pixel as the center, the grayscale difference between it and its neighboring pixels in the four directions (up, down, left, and right) is read. The grayscale difference direction is then vectorized according to the angle between pixels (such as 0°, 90°, 180°, 270°) to construct the grayscale gradient vector field of the tile boundary. After completing the calculation of all boundary pixel vectors, the distribution density and variation frequency of all directional values ​​in the vector field are statistically analyzed to determine the number of grayscale perturbation changes in different directions, generating a directional perturbation frequency sequence. Finally, the sequences are arranged in directional order to form a complete grayscale directional perturbation sequence, which is used to reveal the directional instability distribution of boundary grayscale changes.

[0041] S502: Based on the gray-scale direction perturbation sequence, calculate the difference between the perturbation change frequency and the boundary extension direction, perform interval statistics on the amplitude sequence, filter segments with amplitude within the direction difference threshold, extract the local texture direction information of the segments, and determine whether the texture and brightness change directions are consistent to obtain a set of texture segments with consistent directions. First, identify the angular difference between pixels indicated by two adjacent perturbation directions in the perturbation sequence along the boundary extension direction in the image. Record the angle between each perturbation direction and the boundary direction as the perturbation difference amplitude. Then, assemble all the angle values ​​into a perturbation amplitude sequence. Next, statistically group this sequence according to fixed intervals, for example, 10 degrees as a statistical interval. Divide all amplitude values ​​into segments, count the number of difference amplitudes contained in each interval, and filter out segments with perturbation difference amplitudes less than the threshold according to a set threshold. The directional difference threshold is set based on the image noise range and the allowable directional fluctuation range of the woven texture, generally set to 15 degrees. When the angular differences in a segment are all less than the threshold, it is considered that the directional perturbation of that segment is consistent with the boundary direction. Extract the corresponding texture direction in these segments as the representative direction. Then, compare whether the local texture direction in the direction sequence is consistent with the brightness change direction of the adjacent area. If the difference between the two directions is less than 10 degrees, it is considered that the texture and brightness direction are consistent. Record these texture segments that meet the directional consistency judgment as a set of directional consistent texture segments.

[0042] S503: Based on the set of texture fragments with consistent orientation, call the continuous jump amplitude and boundary gray-scale breakage feature data in the fragments, jointly judge the two, filter the areas where the continuous jump amplitude exceeds the jump baseline threshold and is accompanied by gray-scale breakage, aggregate the spatial index range of the area, and obtain the braided tube appearance defect identification scheme. The algorithm retrieves the internal pixel grayscale information and boundary structure feature data of each segment, extracts the grayscale jump amplitude between consecutive pixels, and performs grayscale abrupt change detection on the boundary region. First, it reads the grayscale value sequence of consecutive pixels in each segment in sequence, and then calculates the grayscale difference between adjacent pixels. If the jump amplitude of three or more consecutive pixels exceeds the jump baseline threshold, it is determined that there is a continuous grayscale disturbance. The jump baseline threshold is set according to the overall texture frequency of the image and the boundary noise control level, and is generally twice the average grayscale change of the boundary region. For example, if the average grayscale change of the boundary region is 7, the jump baseline threshold can be set to 14. During the judgment process, all consecutive pixel segments that meet the jump amplitude exceeding 14 are recorded. These segments are further checked to see if their positions are within the boundary grayscale break region. If a grayscale jump sequence coincides with the grayscale break boundary, it is determined to be an abnormal segment. Finally, all pixel segments that meet the grayscale jump amplitude exceeding the set value and fall within the grayscale break boundary are aggregated into regions. The coordinate range of these regions is extracted and their grid index numbers are sorted to form a complete braided pipe appearance defect identification result.

[0043] Please see Figure 7 A machine vision-based braided tubing appearance defect detection system includes: The brightness break detection module is used to achieve S1: acquiring blue light interference image frames, extracting scan lines along the radial direction of the image center, locating the gray peaks of the scan lines, constructing the brightness change trend through the gray difference of the peaks, extracting the sign jump breakpoints, and generating the brightness break recognition area. The structural anomaly localization module is used to implement S2: select the symmetrical direction scan line in the brightness fracture identification area, calculate the corresponding transverse difference value and dense distribution of the peak, locate the difference accumulation area, map it to the radial yarn interface structure of the braided tube, and generate the structural anomaly area; The burr anomaly detection module is used to implement S3: analyze structurally abnormal regions, extract light spot abrupt change zones and peak extension paths, measure the expansion trend on both sides of the fitted line, screen regions inconsistent with the edge direction, identify continuous morphological jump phenomena, and generate burr anomaly recognition regions. The deformation trend recognition module is used to implement S4: analyze the image grid of the burr anomaly recognition area, extract the gray-scale direction change sequence and perform linear segmentation, count the duration and cross frequency, and generate the deformation recognition area by referring to the diagonal stripe direction as the benchmark. The defect comprehensive identification module is used to implement S5: extract the deformation identification area, measure the gray scale span and disturbance frequency, analyze the area with consistent texture and brightness changes, and combine the boundary gray scale breakage and continuous jump to obtain the braided tube appearance defect identification scheme.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine vision-based method for detecting appearance defects in braided tubing, characterized in that, Includes the following steps: S1: Acquire blue light interference image frames, extract scan lines along the radial direction of the image center, locate the gray peaks of the scan lines, construct the brightness change trend through the gray difference of the peaks, extract the sign jump breakpoints, and generate brightness break recognition areas. S2: Select the symmetrical direction scan line in the brightness fracture identification area, calculate the corresponding transverse difference value and dense distribution of the peak, locate the difference aggregation area, map it to the radial yarn interface structure of the braided tube, and generate the structural abnormal area. S3: Analyze the structurally abnormal region, extract the light spot abrupt change zone and the peak extension path, measure the expansion trend on both sides of the fitted line, screen regions that are inconsistent with the edge direction, identify the phenomenon of continuous morphological jump, and generate burr abnormality identification region. S4: Analyze the burr anomaly identification area, extract the gray-scale direction change sequence and perform linear segmentation, count the duration and cross frequency, and generate the deformation identification area with reference to the diagonal stripe direction as the benchmark. S5: Extract the deformation recognition area, measure the grayscale span and disturbance frequency, analyze the area where the texture and brightness change are consistent, and combine the boundary grayscale break and continuous jump to obtain the braided tube appearance defect recognition scheme.

2. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The brightness breakage identification region includes breakpoint distribution density, grayscale difference sequence, and brightness change trend; the structural anomaly region includes lateral difference value distribution, offset difference aggregation, and misaligned structural bands; the burr anomaly identification region includes spot abrupt change bands, peak extension characteristics, and morphological jump phenomena; the deformation identification region includes grayscale direction continuity length, texture crossover frequency, and directional anomaly segments; and the braided tube appearance defect identification scheme includes texture direction consistent segments, boundary grayscale breakage characteristics, and continuity jump degree.

3. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The grayscale peak of the scan line refers to the pixel position where the grayscale value reaches and is higher than that of the adjacent point along the scan line direction.

4. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The difference clustering area refers to the region where the lateral difference values ​​of corresponding peaks on adjacent scan lines are densely distributed and concentrated in space.

5. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire a blue light interference image frame, extract scan lines divided along the radial direction of the image center, call the gray level in the image frame, locate the local gray maximum value on each scan line, and generate a gray peak localization sequence. S102: Call the grayscale peak positioning sequence, calculate the grayscale difference between adjacent peaks, construct a symbol change sequence, extract the symbol jump position, and generate a jump breakpoint segment index group. S103: Based on the jump breakpoint segment index group, count the number of breakpoints within a unit pixel, determine the breakpoint density distribution, filter out areas where the density exceeds the threshold, and obtain the brightness break recognition area.

6. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Call the scan line located in the symmetrical direction of the image in the brightness fracture recognition area, retrieve the gray peaks with the same number on the left and right scan lines, obtain the coordinate values ​​of the symmetrical peaks on the horizontal axis, calculate the horizontal coordinate difference, and arrange all the differences in order to generate a horizontal difference value sequence. S202: Based on the horizontal difference value sequence, divide the data into segments according to the preset coordinate difference distribution interval, count the number of differences in the interval, determine whether the distribution density exceeds the horizontal difference cluster density threshold, record the corresponding interval index position, and obtain the offset difference cluster index group. S203: Based on the mapping relationship between the offset difference aggregation index group and the image frame to the braided tube structure, retrieve the corresponding radial yarn interface area, mark the area range where there is symmetrical offset, extract the structural boundary information of the area, and obtain the structural abnormal area.

7. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Call the structural anomaly region, extract the gray-level change trajectory in the boundary direction, detect the position of gray-level abrupt change points in the trajectory, extract the brightness enhancement region formed by the continuous arrangement of gray-level abrupt change points, obtain the light spot structure distributed along the edge direction, and establish a light spot abrupt change band path set. S302: Based on the set of light spot abrupt change zone paths, fit the extension direction straight line of each path, and calculate the extension width difference index of the peak distribution on both sides of the path. Make a difference judgment on the left and right extension widths, and filter out the paths whose extension offset direction is not asymmetrical with the edge direction to obtain the region group with inconsistent extension direction. S303: Call the boundary continuous point list of each region in the inconsistent extension direction region group, detect the gray level jump amplitude and jump number of adjacent pixels point by point, determine whether there is continuous fluctuation across the jump amplitude threshold in the gray level response sequence, mark the boundary segment of the region that meets the jump feature, and obtain the spur anomaly identification region.

8. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Call the burr anomaly recognition area, obtain the pixel grayscale values ​​along the row and column directions in each image grid, construct the grayscale change sequence in the horizontal and vertical directions, perform direction extraction operation on the continuous gradient change segments in each group of sequences, divide according to the direction change interval, and obtain the local texture direction sequence set; S402: Based on the local texture direction sequence, the duration of each direction change and the number of times the intersection direction appears are counted to form the direction change feature vector of each grid. The direction information of the diagonal cross area on the surface of the braided tube is used as a reference to calculate the direction difference value between the feature vector and the reference direction and obtain the texture direction difference matrix. S403: Based on the distribution of orientation differences in the grids in the texture orientation difference matrix, determine whether the orientation change exceeds the orientation difference threshold in the continuous grids, mark the continuous segments of abnormal changes, extract the index of the covered image region, and obtain the deformation recognition region.

9. The method for detecting appearance defects in braided tubing based on machine vision according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Call the edge image of the patch in the deformation recognition area, extract the set of boundary pixels of each patch, detect the gray value of the boundary pixels, calculate the gray range span index, and construct a gray gradient vector field in combination with the distribution direction of the boundary pixels. Measure the directional perturbation frequency of the vector field and obtain the gray direction perturbation sequence. S502: Based on the gray-scale direction perturbation sequence, calculate the difference between the perturbation change frequency and the boundary extension direction, perform interval statistics on the amplitude sequence, filter segments with amplitude within the direction difference threshold, extract the local texture direction information of the segments, and determine whether the texture and brightness change direction are consistent to obtain a set of texture segments with consistent direction. S503: Based on the set of texture fragments with consistent direction, call the continuous jump amplitude and boundary gray-scale breakage feature data in the fragments, perform joint judgment on the two, filter out the regions where the continuous jump amplitude exceeds the jump baseline threshold and is accompanied by gray-scale breakage, aggregate the spatial index range of the regions, and obtain the braided tube appearance defect identification scheme.

10. A machine vision-based system for detecting appearance defects in braided tubing, characterized in that, The system is used to implement the machine vision-based braided tube appearance defect detection method according to any one of claims 1-9, the system comprising: The brightness break detection module is used to achieve S1: acquiring blue light interference image frames, extracting scan lines along the radial direction of the image center, locating the gray peaks of the scan lines, constructing the brightness change trend through the gray difference of the peaks, extracting the sign jump breakpoints, and generating the brightness break recognition area. The structural anomaly localization module is used to implement S2: select the symmetrical direction scan line in the brightness fracture identification area, calculate the corresponding transverse difference value and dense distribution of the peak, locate the difference accumulation area, map it to the radial yarn interface structure of the braided tube, and generate the structural anomaly area; The burr anomaly detection module is used to achieve S3: analyze the structural anomaly region, extract the light spot abrupt change zone and peak extension path, measure the expansion trend on both sides of the fitted line, screen regions that are inconsistent with the edge direction, identify continuous morphological jump phenomena, and generate burr anomaly recognition region. The deformation trend recognition module is used to implement S4: analyze the image grid of the burr anomaly recognition area, extract the gray-scale direction change sequence and perform linear segmentation, count the duration and cross frequency, and generate the deformation recognition area by referring to the diagonal stripe direction as a reference. The defect comprehensive identification module is used to implement S5: extract the deformation identification area, measure the gray scale span and disturbance frequency, analyze the area with consistent texture and brightness changes, and combine the boundary gray scale breakage and continuous jump to obtain the braided tube appearance defect identification scheme.

Images