Connector wire harness crimping defect detection method based on image recognition

Abstract

The invention relates to the technical field of defect detection, in particular to a connector wire harness crimping defect detection method based on image recognition, which comprises the following steps of: detecting the crimping defect of a connector wire harness based on image recognition; the method comprises the following steps: segmenting a crimping region image, extracting local brightness distribution characteristics, analyzing abnormal aggregation, detecting a port bare signal and contact uniformity, judging in combination with an area ratio, identifying a wiring harness crimping defect, and outputting a risk index. According to the method, image segmentation and regional brightness contrastive analysis are adopted, the recognition capability of micro-differences of all structures in a crimping region is refined, and a discrimination mechanism of spatial distribution anomaly aggregation is combined, so that hierarchical distinguishing of three-dimensional structure deformation, hidden defects and local anomalies, multi-stage reflection feature sorting and regional mutation detection are realized; the fine difference between metal exposure and coating material transition areas is effectively captured, and the contact distribution and stress consistency of the wire crimping part is reflected by combining pixel-level path tracking and gray uniformity comparison.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method for detecting connector harness crimping defects based on image recognition. Background Art

[0002] Defect detection technology is an important branch of intelligent manufacturing and quality control. It is widely used in industrial production for the automatic identification and evaluation of product appearance, structural integrity and functional status. This technology uses sensors, image acquisition devices (such as industrial cameras), image processing algorithms and artificial intelligence models to monitor products in real time and identify various defects such as cracks, deformations, foreign matter, missing parts, stains, etc. In recent years, with the development of computer vision and deep learning, the application of image recognition in defect detection has gradually become popular, improving the automation, accuracy and efficiency of detection, reducing the burden of manual inspection and improving production yield, becoming a key quality assurance method in the intelligent manufacturing system.

[0003] Among them, the image recognition connector harness crimping defect detection method is a quality inspection solution that applies image recognition technology to the connector harness crimping process in the automotive, electronics, electric power and other industries. Its purpose is to use high-resolution image acquisition devices and intelligent recognition algorithms to automatically detect the wire harness after crimping to determine whether there are defects such as poor crimping, deformation, broken strands, skewness, and false connections. This technology improves the accuracy and speed of detection, reduces human error, ensures the electrical performance and mechanical stability of key components, and is an important means to ensure product safety and reliability.

[0004] Existing technologies generally use image detection from a single angle or fixed area, and the processing flow is limited to overall brightness statistics or regional mean judgment, which makes it difficult to reveal complex anomalies in structural hierarchy and spatial distribution. There is a lack of linkage analysis for three-dimensional distortion, local exposure and multi-layer spatial heterogeneity, and it is impossible to make precise judgments on local spatial fluctuations or microscopic variations. The detection results are highly dependent on empirical interpretation or template comparison, which makes it difficult to adapt to diverse product structures and complex manufacturing scenarios. There is a lack of independent feature quantification support for segmented risks such as path distribution and contact uniformity, which makes it easy to misjudge local defects and fuzzy risk classification, affecting the self-monitoring capability and quality of the production process. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a connector harness crimping defect detection method based on image recognition.

[0006] In order to achieve the above-mentioned object, the present invention adopts the following technical solution: a method for detecting connector harness crimping defects based on image recognition, comprising the following steps:

[0007] S1: Based on the image data of the connector harness crimping area, it is divided into multiple grids, the grayscale brightness mean of the edge and center pixels of each grid is compared to determine the local brightness distribution, and the grid brightness density gradient is integrated to obtain the local brightness distribution characteristics;

[0008] S2: Based on the local brightness distribution characteristics, calculate its distribution in the crimping area of ​​the crimping terminal, analyze the normalized performance of the grid brightness density gradient, identify the spatial deviation aggregation area, determine its spatial correspondence, and obtain a three-dimensional abnormal aggregation factor;

[0009] S3: Based on the three-dimensional abnormal aggregation factor, analyze the reflection characteristics of the image of the harness port area under the target light source, optimize the pixel reflection intensity sorting, screen the reflection intensity mutation area, compare the light energy density of the mutation area and the coating material area, and obtain the port exposure characteristic signal;

[0010] S4: Based on the exposed port characteristic signal, determine the distribution of the wire path in the crimping terminal contact area image, analyze the wire pixel extension distribution, compare the grayscale uniformity of the contact area pixels, calculate the grayscale distribution of the contact area, and obtain the crimping terminal contact uniformity coefficient.

[0011] The improvements of the present invention are that the local brightness distribution characteristics include brightness change characteristics, distribution uniformity, and structural clarity; the three-dimensional abnormal aggregation factor includes the number of abnormal areas, the degree of aggregation distribution, and the range of connected areas; the port exposure characteristic signal includes reflection intensity characteristics, metal exposure marks, and signal change types; the indenter contact uniformity coefficient includes contact area distribution, coverage consistency, and grayscale uniformity.

[0012] The present invention is improved in that the step of acquiring the local brightness distribution feature is specifically as follows:

[0013] S111: Based on the image data of the crimping area of ​​the connector harness, the crimping area image is divided into a plurality of square grid areas, and a difference analysis is performed by comparing the grayscale brightness mean of each grid edge pixel with the center pixel to obtain a pixel brightness difference set;

[0014] S112: Calculate the absolute value of the brightness mean difference of each grid based on the pixel brightness difference set, filter out grid areas that are greater than the brightness density standard, count the number of selected grids, and compare their distribution in the image space to obtain local brightness distribution characteristics.

[0015] The present invention is improved in that the steps of obtaining the three-dimensional abnormal aggregation factor are specifically as follows:

[0016] S211: Based on the local brightness distribution characteristics, calculate the difference between the edge and center brightness mean of each grid in the crimping area of ​​the crimping terminal, normalize the difference, and locate the brightness distribution of each grid in three-dimensional space to obtain grid brightness offset distribution data;

[0017] S212: Based on the grid brightness offset distribution data, determine the continuous change of the normalized gradient of adjacent grids, identify the formed clustered areas, and perform connectivity detection to obtain spatial offset clustering performance;

[0018] S213: According to the spatial offset aggregation performance, compare the aggregation center with the area corresponding to the boundary of the crimping area of ​​the crimping terminal. If the spatial position has an overlap, inclusion or adjacency relationship, mark the aggregation area as an abnormal aggregation area to obtain a three-dimensional abnormal aggregation factor.

[0019] The present invention is improved in that the step of obtaining the port exposure characteristic signal is specifically as follows:

[0020] S311: Based on the three-dimensional abnormal aggregation factor, its regional distribution is analyzed, the order of pixel reflection intensity of the grid in the harness port area is optimized, the grayscale sequence change trend is calculated, and the sections with continuous drastic grayscale changes are screened to obtain a continuous mutation grayscale interval group;

[0021] S312: comparing the average reflection intensity of the pixel set in the local space corresponding to the continuous mutation grayscale interval group, calculating the difference with the average reflection intensity of the coating material area, adjusting the proportional influence on the number of pixels, and obtaining the reflection intensity difference;

[0022] S313: Analyze the pixel distribution in the reflection abnormality signal domain based on the reflection intensity difference, determine the spatial aggregation state, optimize the boundary consistency of the continuous abnormal area, compare the reflection characteristics with the metal area, identify the boundary features, and obtain the port exposure feature signal.

[0023] The present invention is improved in that the steps for obtaining the indenter contact uniformity coefficient are specifically as follows:

[0024] S411: Based on the exposed port characteristic signal, analyzing the pixel coordinate distribution of the wire path in the contact area image, calculating the grayscale data sequence of each pixel on the path, unifying the grayscale distribution range through normalization conversion, and comparing the spatial variation of grayscale between different pixels to obtain the wire grayscale extension distribution;

[0025] S412: Calling the grayscale extension distribution of the wire, determining the distribution state of the standardized path grayscale sequence in the contact area, calculating the grayscale balance level in the area, analyzing the discrete relationship between the path grayscale and the balance level, optimizing the grayscale difference combination, identifying the optimal fluctuation amplitude, and obtaining the grayscale discrete amplitude of the contact area;

[0026] S413: Based on the grayscale discrete amplitude of the contact area, calculate the ratio to the path grayscale balance level, calculate the grayscale distribution of the contact area, and obtain the indenter contact uniformity coefficient.

[0027] The present invention is improved in that the steps further include:

[0028] S5: Based on the indenter contact uniformity coefficient, calculate the correspondence with the wire extension length, analyze the ratio of the number of pixels in the contact area to the path extension area, determine whether the ratio is in a balanced range, identify the unbalanced endpoints, and obtain a connection balance risk index;

[0029] The connection balance risk indicators include risk level, number of abnormal endpoints, and balance analysis results.

[0030] The present invention is improved in that the step of obtaining the connection balance risk indicator is specifically as follows:

[0031] S511: Based on the indenter contact uniformity coefficient, the grayscale uniformity in the indenter contact area image is analyzed. In combination with the contact area distribution and coverage consistency, the number of effective contact pixels within the wire path is calculated, and the pixel distribution within the area is determined to obtain the contact pixel density.

[0032] S512: Based on the contact pixel density, the ratio of the number of effective contact pixels to the extended area of ​​the wire path is compared with the extended area of ​​the wire path, and the distribution consistency of the pixels in the contact area is optimized to obtain a contact balanced distribution feature.

[0033] S513: Based on the contact balance distribution characteristics, identify the wiring harness endpoints that do not meet the balance requirements, determine the distribution status of the associated wire paths, and obtain a connection balance risk index based on the wiring harness number and distribution characteristics.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are:

[0035] In the present invention, by adopting image segmentation and regional brightness contrast analysis, the recognition ability of micro-differences of various structures inside the crimping area is refined, and the discrimination mechanism of spatial distribution abnormal aggregation is combined to realize the hierarchical distinction of three-dimensional structural deformation, hidden defects and local abnormalities. Multi-stage reflection feature sorting and regional mutation detection are used to effectively capture the subtle differences in the transition area between metal exposure and coating materials. Combined with pixel-level path tracing and grayscale uniformity comparison, the contact distribution and force consistency of the wire crimping part are reflected. By using mutual confirmation and multi-parameter linkage analysis, the limitation of single feature parameters is broken through, the adaptability to complex structural states is improved, and hierarchical and partitioned intelligent evaluation of defect risks is realized, which facilitates automatic screening and accurate classification of wire harnesses in various scenarios and improves the accuracy of defect positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the main steps of the present invention;

[0037] Figure 2 Flowchart for obtaining local brightness distribution characteristics in the present invention;

[0038] Figure 3 This is a flow chart for obtaining the three-dimensional abnormal aggregation factor in the present invention;

[0039] Figure 4 This is a flow chart for obtaining the port exposure characteristic signal in the present invention;

[0040] Figure 5 This is a flow chart for obtaining the pressure head contact uniformity coefficient in the present invention;

[0041] Figure 6 This is a flow chart for obtaining the connection balance risk indicator in the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0043] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0044] Example

[0045] See also Figure 1 The present invention provides a technical solution: a method for detecting connector harness crimping defects based on image recognition, comprising the following steps:

[0046] S1: Based on the image data of the connector harness crimping area, the image is segmented into multiple grid areas. The grayscale brightness mean of the edge pixels and the center pixels in each grid are compared to determine the distribution characteristics of the brightness of each grid. The brightness density gradient of the grid is then integrated to obtain the local brightness distribution characteristics.

[0047] S2: Based on the local brightness distribution characteristics, its distribution in the crimping area of ​​the crimping terminal is calculated, and the normalized performance of the brightness density gradient of each grid is analyzed. The clustered areas that deviate from the crimping center position in space are identified, and the corresponding relationship between their distribution and the crimping area of ​​the crimping terminal is determined to obtain the three-dimensional abnormal clustering factor;

[0048] S3: Based on the three-dimensional abnormal aggregation factor, the reflection characteristics of the image of the harness port area under the illumination of the target light source are analyzed. The sorting method of the pixel reflection intensity of each grid area is optimized. The local areas with continuous reflection intensity mutations are screened. The light energy density performance of the mutation area and the coating material area is compared to obtain the port exposure characteristic signal;

[0049] S4: Based on the exposed port feature signal, determine the distribution of the wire path in the image of the crimping terminal contact area, analyze the extended distribution of the wire pixels in the path, compare the uniformity of the pixel grayscale distribution in the contact area, calculate the grayscale distribution of the contact area, and obtain the crimping terminal contact uniformity coefficient;

[0050] S5: Based on the contact uniformity coefficient of the indenter, calculate the correspondence between it and the wire extension length, analyze the area ratio of the number of pixels in the contact area and the path extension, determine whether the area ratio is in the contact balance range, identify the wire harness endpoints with insufficient contact balance, and obtain the connection balance risk index.

[0051] The local brightness distribution characteristics include brightness change characteristics, distribution uniformity, and structural clarity. The three-dimensional abnormal aggregation factor includes the number of abnormal areas, the degree of aggregation distribution, and the range of the connected area. The port exposure characteristic signal includes the reflection intensity characteristics, metal exposure mark, and signal change type. The indenter contact uniformity coefficient includes the contact area distribution, coverage consistency, and grayscale uniformity. The connection balance risk indicator includes the risk level, the number of abnormal endpoints, and the balance analysis results.

[0052] In S1, the grid area refers to the division of the crimping area image into several small, regularly arranged square or rectangular areas according to certain rules. The purpose of this is to localize the information of the overall image and facilitate data comparison and feature analysis at a more detailed spatial level; the edge pixel refers to the pixel points in the outermost circle (close to the grid boundary) of each grid area. The pixels are usually used to reflect the information of the regional outline or structural boundary; the center pixel refers to the pixel point close to the center position inside each grid area. The brightness information of the center pixel can better reflect the main structural features inside the grid; the brightness density gradient refers to the difference between the brightness mean of the edge pixel and the center pixel inside each grid. This parameter reflects the intensity and trend of the brightness distribution change in the local area, which helps to determine whether there are three-dimensional structural abnormalities such as indentations and bulges in the area. In S2, the deviated clustered area refers to the area where the brightness characteristics of some grids are significantly different from the overall average state, and the abnormal grids show a continuous and clustered distribution feature in space by performing spatial statistics on the brightness density gradients of all grids. The abnormal clustering phenomenon usually indicates that there is a three-dimensional structural abnormality at that location; the corresponding relationship refers to the spatial relationship between the above-mentioned abnormal clustering area and the crimping area of ​​the crimping terminal, such as overlap, inclusion or adjacency in spatial position, which is used to determine whether the detected abnormal area is located in the actual crimping position; the three-dimensional abnormal clustering factor is a criterion or parameter that characterizes the risk of three-dimensional structural abnormality, calculated based on the spatial distribution and brightness changes of the clustering area, which comprehensively reflects the degree of spatial abnormality aggregation and distribution position, and is usually a numerical or factor output for subsequent risk judgment. In S3, reflection characteristics refer to the reflection strength and uniformity of light on the surfaces of different materials in the harness port area under the illumination of a specific light source, which is often reflected through the pixel intensity information of the image; the sorting method refers to the processing method of arranging the reflection intensity data of all pixels in a certain area or multiple areas from high to low or from low to high, which is convenient for discovering strong and weak mutation points; the local area refers to a small area composed of pixels with similar spatial positions and adjacent to each other within the overall port image or a specific grid range, which is convenient for detecting reflection characteristic mutation points or abnormal distribution; the light energy density performance refers to the overall distribution of the reflection intensity of all pixels in a small area, which is the basis for determining the differences in materials such as bare metal and insulation coating. In S4, the wire path refers to the pixel connection line in the image starting from the point where the wire enters the crimping end and traces along the direction of the wire to the end of the crimping area, reflecting the actual distribution and direction of the wire; the extended distribution refers to the total number of pixels spanned by the wire path in the image and its spatial distribution pattern, which is used to describe the extension length and coverage of the wire in the crimping area; the uniform feature refers to the statistical distribution of the grayscale (brightness) values ​​of all pixels in the contact area of ​​the indenter to determine whether its grayscale distribution is obviously concentrated or dispersed, which is used to evaluate the force or contact consistency of the contact surface; the indenter contact uniformity coefficient is an indicator derived from the statistical characteristics of the grayscale distribution, which is used to measure the uniformity of the force or contact distribution on the contact surface of the indenter area.This coefficient can be obtained through mathematical methods such as pixel grayscale uniformity and variance. In S5, the wire extension length refers to the length covered by the wire path in the image through pixel tracking (i.e., the number of pixels along the wire path from the point of entry to the crimping area), which is used to evaluate whether the wire has fully entered the crimping area; the area ratio refers to the proportional relationship between the number of effective contact pixels in the contact area and the wire path extension length, reflecting the relative relationship between the actual contact area and the theoretical space between the wire and the crimping head; the contact balance interval refers to the area ratio interval preset based on statistics or process standards, which is used to distinguish the boundary between normal contact and poor contact; the contact balance comprehensively evaluates the force, coverage, and contact distribution of the wire in the crimping end area, which is used to determine the reliability level of the contact quality.

[0053] See also Figure 2 , the steps for obtaining local brightness distribution features are as follows:

[0054] S111: Based on the image data of the crimping area of ​​the connector harness, the crimping area image is divided into a plurality of square grid areas, and a difference analysis is performed by comparing the grayscale brightness mean of each grid edge pixel with the center pixel to obtain a pixel brightness difference set;

[0055] First, the crimping area in the center of the image is extracted and divided into several 40×40 pixel square grid areas according to the equal division rule. The outermost circle of pixels in each grid is extracted as edge pixels, and a set of pixels within a 5×5 pixel range at the center is extracted as the center pixel. After obtaining the grayscale value of each pixel, the average grayscale calculation operation is performed on the edge pixels and the center pixels respectively to form the edge mean and center mean of the grid. The absolute difference between the two is then calculated as the brightness difference index. The above steps are repeated until all grids are processed and a brightness difference set is formed. Then, the difference value is analyzed one by one to see if it exceeds the set brightness density judgment standard. The judgment standard value is set by the previous sample statistics method and the value is 12. Grids with values ​​higher than 12 are marked as areas with significant brightness differences, and the grid numbers are recorded. Subsequently, the number of all marked grids is counted and their positions in the image space are extracted. The position information is represented in the form of grid row and column coordinates. A two-dimensional distribution map of the abnormal grids in the image space is further drawn to generate the local brightness distribution characteristics of the current image.

[0056] S112: Calculate the absolute value of the brightness mean difference of each grid based on the pixel brightness difference set, select grid areas with a brightness density greater than a standard, count the number of selected grids, and compare their distribution in the image space to obtain local brightness distribution characteristics;

[0057] The difference value of each grid is processed one by one, and its absolute value is taken and compared with the set brightness density standard value to determine whether it meets the condition greater than the threshold. The grids that meet the conditions are screened and their corresponding numbers and positions are recorded. At the same time, the position of the grid is converted into a set of image coordinate points, and then the total number of selected grids on the image is counted. At the same time, a coordinate set is established for subsequent spatial distribution evaluation. In the distribution analysis process, by observing whether the abnormal grid coordinate points are concentrated in a certain area of ​​the image, it is determined whether a clustering effect is formed. For example, if there are a large number of continuous abnormal grids in the central area of ​​a certain image, it is regarded as a local structural abnormality, and the number of abnormal grid concentration areas is further counted to determine whether it is close to the center of the crimping area. If the concentrated area is mainly distributed at the edge of the image, it is determined to be affected by the non-crimping part. Otherwise, it is identified as a structural abnormality-related area. The spatial distribution information is used to determine whether there is an abnormal concentration area of ​​brightness difference in the image, and it is output as a local brightness distribution feature.

[0058] See also Figure 3 , the specific steps for obtaining the three-dimensional abnormal aggregation factor are:

[0059] S211: Based on the local brightness distribution characteristics, calculate the difference between the average brightness of the edge and the center of each grid in the crimping area of ​​the crimping terminal, normalize the difference, and locate the brightness distribution of each grid in three-dimensional space to obtain grid brightness offset distribution data;

[0060] First, each grid area within the crimping area of ​​the crimped terminal is read sequentially. For each grid, edge pixels and center pixels are extracted, and their grayscale brightness values ​​are obtained and averaged. The two means are then subtracted and their absolute value is taken. The brightness difference value of each grid is recorded. After processing all grids, the brightness difference of each grid is linearly converted to a standardized interval based on the maximum and minimum values ​​of all brightness differences in the entire image. Three decimal places are retained during the conversion process. This is used as the height index for subsequent three-dimensional positioning. The normalized result is then combined with the two-dimensional image coordinates of the corresponding grid to form a set of three-dimensional description points. All three-dimensional points are then arranged into a brightness offset model based on the row and column order of the grids in the image. This model reflects the degree of grayscale difference of each grid relative to the center of the overall crimping area. For example, if an image of size 300×300 pixels is processed and divided into 49 grids, and the brightness difference values ​​of multiple consecutive grids in a certain area are high and concentrated to the left of the image center, a local block with continuous high height will be formed in the three-dimensional structure. This three-dimensional block is used to determine whether there is concentrated and uneven brightness distribution in the crimping area and obtain the grid brightness offset distribution data.

[0061] S212: Based on the grid brightness offset distribution data, determine the continuous change of the normalized gradient of adjacent grids, identify the formed clustered areas, and perform connectivity detection using the formula:

[0062]

[0063] Obtain the spatial offset aggregation performance GU, which is used to measure the spatial offset and aggregation characteristics of the aggregation area, where ΔLU i Represents the brightness gradient of the i-th grid, the normalized brightness difference between the edge and the center, which is used to reflect the change in local brightness distribution. i Represents the area of ​​the i-th grid, which measures the coverage space of the grid in the image, d j It represents the spatial distance between the jth grid in the clustering area and the crimping center, and is used to reflect the distribution breadth of each grid in the clustering area to the crimping center. k Represents the coordinates of the brightness center of the kth grid in the clustering area, representing the center position of the brightness distribution inside the grid (i.e., the local brightness center of gravity). is the mean of the brightness centroid coordinates of all grids in the clustering area, reflecting the center of the overall brightness distribution (global brightness centroid), n represents the number of grids involved in the calculation, m represents the number of grids used to calculate the spatial distance in the clustering area, and q represents the number of grids used to calculate the brightness centroid coordinates in the clustering area;

[0064] To determine the continuous change of the normalized gradient of adjacent grids, first calculate the difference in the normalized gradient between each two adjacent grids in the order of grid numbers and make a positive or negative direction judgment. If the gradient differences of three or more consecutive grids are all increasing or decreasing in the same direction, they are marked as the initial clustering area. If the normalized brightness gradients of grids 1, 2, and 3 are 1.05, 1.22, and 1.38 respectively, the adjacent differences are 0.17 and 0.16, both in the positive direction, and the three constitute a continuously changing area. Then, a connectivity check is performed to exclude grids with number jumps or discontinuous spatial distributions. For example, if the number between grids 3 and 5 is missing or the spatial distance is greater than the regular spacing of the image grid distribution, it is determined to be a non-connected area and this part is eliminated. After obtaining a continuous and adjacent clustering area, the normalized brightness gradient ΔLU of each grid in the area is extracted in turn. i , corresponding area AU i , the spatial Euclidean distance d from the crimping center j , and its brightness center coordinate CU k , then introduce the formula to calculate the spatial offset aggregation performance. The numerator is used to measure the cumulative product between the total brightness change and the corresponding area coverage area. The first term of the denominator reflects the geometric dispersion of the aggregation area. The square root of the square distance from all grids to the center of the crimping is calculated to obtain the spatial breadth. The second term is the average of the brightness center of each grid and the brightness center of all grids in the aggregation area. The sum of squared deviations between them is used to measure the concentration of the brightness center of gravity within the clustered area. For four grids in a clustered area, their normalized brightness gradients are 12.5, 9.7, 15.3, and 10.8, respectively, and each area is 1.2 square millimeters. The molecular part is 12.5×1.2+9.7×1.2+15.3×1.2+10.8×1.2=57.96.

[0065] The spatial distances between each grid and the crimping center are 2.5mm, 3.1mm, 2.9mm, and 3.3mm respectively. The square root of the square root of the square root is

[0066] The brightness centers are: 8.2, 7.9, 8.4, 8.0

[0068] The mean is:

[0069]

[0070] The sum of the squares of the deviations of the centers of gravity is:

[0071] (8.2-8.125) 2 +(7.9-8.125) 2 +(8.4-8.125) 2 +(8.0-8.125) 2 =0.1474;

[0072] The denominators total:

[0073] 5.93+0.1474≈6.08;

[0074] Calculated:

[0075]

[0076] The larger the numerical value of the calculation result, the stronger the spatial concentration and brightness difference of the aggregation area, and the spatial offset aggregation performance is obtained. The formula constructs a quantitative model of the image spatial structured light distribution aggregation through the multi-factor superposition of brightness gradient, grid area, spatial distance and brightness center of gravity deviation, thereby effectively expressing the structural offset phenomenon existing in the crimping area image.

[0077] S213: Based on the spatial offset clustering performance, the clustering center is compared with the area corresponding to the boundary of the crimping area of ​​the crimping terminal. If the spatial positions overlap, contain, or are adjacent, the clustering area is marked as an abnormal clustering area, and a three-dimensional abnormal clustering factor is obtained.

[0078] First, all the three-dimensional points of the grids in the aforementioned brightness offset model are screened, and those grids with brightness differences greater than the median level are selected as candidate abnormal points. Then, the candidate points are analyzed to see whether they form a dense distribution area in the image space. In the dense recognition process, the image coordinates of the candidate points are compared one by one to determine whether the adjacent grids are arranged continuously in the horizontal or vertical direction. If there are more than three adjacent grids that meet this condition in any direction, they are divided into a cluster block. Next, the average position of each cluster block is obtained as the centroid coordinate of the cluster block, and then compared with the boundary area of ​​the crimping area of ​​the crimping terminal. The area is predefined as the central part of the image, and it is judged whether the center of gravity of the cluster block is within the boundary. If it is within the boundary, it is judged to be overlapping. If any corner point outside the cluster block is within the boundary, it is considered to be contained. If the distance between the cluster block and the boundary in the image does not exceed 10 pixels, the two are considered to be adjacent. In the actual detection process, if the center of gravity of a cluster block in the image is near the crimping boundary and there are four consecutive high-brightness grids arranged in the lower right area of ​​the image, then the block meets the abnormal cluster area judgment standard. Finally, all cluster blocks that meet the conditions are numbered and marked, and summarized as a three-dimensional abnormal clustering factor.

[0079] See also Figure 4 ,The specific steps for obtaining the port exposure characteristic signal are:

[0080] S311: Based on the three-dimensional abnormal clustering factor, its regional distribution is analyzed, the order of pixel reflection intensity of the grid in the harness port area is optimized, the grayscale sequence change trend is calculated, and the sections with continuous drastic grayscale changes are screened to obtain a group of continuous mutation grayscale intervals;

[0081] First, the two-dimensional spatial coordinate range corresponding to the marked abnormal area in the image is extracted and mapped to the grids divided in the harness port area. Then, the reflection intensity information of each pixel point is extracted in the grid in turn, and the grayscale channel value is called as the single-pixel reflection intensity reference value. All pixels are sorted from low to high according to the grayscale value to form a grayscale sequence. During the sorting process, the image position index of the corresponding pixel is retained for subsequent positioning, and the change trend calculation operation is continued. The adjacent grayscale difference sequence is calculated in turn through the grayscale difference of adjacent pixels, and the segments with continuous difference values ​​greater than a certain threshold are identified. The grayscale difference threshold is set to 20, which is obtained by measuring the insulation coating. The average value of the grayscale difference from the metal reflection is set, and then the start and end positions of the continuous grayscale mutation segments are screened. If the grayscale mutation continues for more than 10 pixels, it is recorded as a continuous grayscale drastic change segment. When processing an image with a resolution of 300×300, there is a grayscale jump from 50 to 200 between the 25th to 40th sorted pixels and the jump trend remains greater than 20, which meets the mutation judgment standard. The system saves its start and end positions and numbers it as grayscale mutation segment 1. Repeat the operation until all pixel grayscale data in all grids are processed. Finally, all segments that meet the grayscale mutation judgment standard are combined into a continuous mutation grayscale interval group.

[0082] S312: Compare the average reflection intensity of the pixel set in the local space corresponding to the continuous mutation grayscale interval group, calculate the difference with the average reflection intensity of the coating material area, and adjust the proportional impact on the number of pixels using the formula:

[0083]

[0084] Get the reflection intensity difference, where SQ z represents the reflection intensity difference of the z-th local spatial region, RQ zo is the reflection intensity of the o-th pixel in the z-th area, AQ z Indicates the coverage area of ​​all pixels in the zth region, MQ zy is the reflection intensity of the yth cladding material pixel in the zth area, BQ z Indicates the coverage area of ​​the pixel points of the coating material in the zth region, NQ z Indicates the number of all pixels in the zth region, PQ z Indicates the number of pixels of the coating material in the zth area;

[0085] Obtain the sum of the reflection intensities of all pixels in each local area and the coverage area, divide the two to obtain the average reflection intensity of the area, and at the same time obtain the sum of the reflection intensities of the coating material pixels in the same area and the coating material coverage area, divide them to obtain the average reflection intensity of the coating material area, then take the absolute value of the difference between the two and multiply it by the square root of the ratio of the total number of pixels in the area to the number of coating material pixels. Using the formula, taking area Z1 as an example, let:

[0086] The total reflection intensity of all pixels is 9500cd / m 2 , the number of pixels is 600, and the coverage area is 120mm 2 ;

[0087] The total reflection intensity of the covering material is 6200cd / m 2 , the number of pixels is 400, and the coverage area is 80mm 2 .

[0088] but:

[0089]

[0090] This value indicates that within the Z1 region, there is a difference of approximately 2.05 in the average reflection intensity per unit area between the overall pixel set and the pixels of the coating material. Based on the standard reflection performance reference of optical materials, if the average reflection difference of the bare metal area exceeds 1.8, it has a strong indicative significance. Therefore, this area can be judged as an abnormal reflection area and enter the next stage of judgment. The formula introduces the average reflection intensity difference, area normalization, and the square root factor of the pixel density ratio to effectively distinguish areas with similar grayscale but different distribution density, thereby enhancing the recognition sensitivity of the bare metal area and obtaining the reflection intensity difference.

[0091] S313: Analyze the pixel distribution within the reflection abnormality signal domain based on the reflection intensity difference, determine the spatial aggregation state, optimize the boundary consistency of the continuous abnormal area, compare the reflection characteristics with the metal area, identify the boundary features, and obtain the port exposure feature signal;

[0092] First, traverse the identified mutation grayscale interval group, relocate the pixel positions contained in each interval to the corresponding coordinates in the original image, and count their spatial distribution density in the image plane. Then, perform aggregation state analysis on the pixel set corresponding to each interval. If the number of aggregation points in a certain window area exceeds 20, it is judged that there is abnormal aggregation, and continue to perform boundary consistency optimization processing. Extract boundary pixels at the edge of each abnormal area, compare the difference between their grayscale values ​​and the center grayscale point by point, and if the difference between the boundary grayscale and the center in a certain direction is too small, expand the boundary in that direction outward by 5 pixels and recalculate the boundary range to ensure The grayscale difference trend in the boundary area is consistent. On this basis, the preset reflection grayscale value range of the metal area is called. This range is taken from the standard metal wire harness head sample, and the reflection intensity range is 180 to 255. Each optimized abnormal boundary area is compared with this grayscale range pixel by pixel. If at least 70% of the pixel grayscales in the area are within this range, it is judged to have metal reflection characteristics. Finally, all abnormal areas that meet the three conditions of aggregation, boundary consistency and metal reflection are located as port exposed sections. Each section is assigned a unique identification number and its image coordinates and grayscale distribution range are recorded. After aggregation, the port exposed feature signal is formed.

[0093] See also Figure 5 , the steps for obtaining the indenter contact uniformity coefficient are as follows:

[0094] S411: Based on the exposed port feature signal, the pixel coordinate distribution of the wire path in the contact area image is analyzed, the grayscale data sequence of each pixel on the path is calculated, the grayscale distribution range is unified through normalization conversion, and the spatial variation of the grayscale between the different pixels is compared to obtain the grayscale extension distribution of the wire;

[0095] First, the image coordinate range corresponding to the marked exposed area is extracted from the crimped end image. After determining the entry path of the wire, the grayscale value is scanned pixel by pixel along the path direction, and the image coordinates and grayscale value of each pixel are recorded and combined to form a grayscale sequence. After the original grayscale sequence is constructed, a standardization conversion operation is performed to linearly stretch the grayscale values ​​of all pixels and uniformly convert them to the range of 0 to 1. After the conversion, the grayscale trend analysis of the pixels on the entire path is performed to determine the continuous change amplitude of the grayscale value on the path, identify the area with a large grayscale jump amplitude and record its jump start and end positions, and extract the pixel positions between the jump points to calculate the overall grayscale value in the segment. The change range and the mean of single-step change are calculated. If the single-step change value continuously exceeds 0.1 and the change continues for more than 20 pixels, the segment is marked as a segment with drastic grayscale change. The difference comparison between adjacent grayscale change values ​​is continued to screen out the grayscale segment with the fastest change speed. Finally, the data of all change segments are combined and combined in the order of their paths to form the wire grayscale extension distribution data. If a path consisting of 120 pixels is extracted from the image, where the grayscale of the 45th to 75th pixels increases from 0.2 to 0.9 and the change value in each step exceeds 0.1, the area is marked as a high-intensity extension area. Finally, the high-frequency change information is collected and output as the wire grayscale extension distribution.

[0096] S412: Call the wire grayscale extension distribution, determine the distribution state of the standardized path grayscale sequence in the contact area, calculate the grayscale balance level in the area, analyze the discrete relationship between the path grayscale and the balance level, optimize the grayscale difference combination, identify the optimal fluctuation amplitude, and obtain the grayscale discrete amplitude of the contact area;

[0097] The grayscale sequence data is mapped to the spatial coordinates of the contact area of ​​the image, and the range of the area is delineated. The distribution density of the path grayscale value in the contact area is counted. In the distribution process, every 10 pixels are used as a sliding window, and the standard deviation of all grayscale values ​​in the window is calculated as the local balance level indicator. After repeating this process to cover the entire area, the standard deviation values ​​in all sliding windows are constructed into a set of grayscale balance data sequences. Then, the difference range between the maximum and minimum values ​​in the balance sequence is analyzed. If the difference between the maximum and minimum values ​​exceeds 0.4, it means that there is obvious grayscale fluctuation in the path. On this basis, the grayscale balance data and the original path grayscale sequence are calculated. The mean difference between the two sets of data is calculated. If the difference in the degree of discreteness of the two sets of data is greater than 0.2, it is considered that there is an unstable grayscale distribution area in the path. Then, the mean comparison before and after is performed on each grayscale mutation position. The point sets with prominent grayscale changes are selected, merged into continuous change segments and sorted. Finally, the grayscale jump segments with the most obvious change trend, largest amplitude, and most concentrated range are output as the optimal fluctuation amplitude item. Combined with the actual image measurement case, it is found that the sliding window standard deviation of a path fluctuates between 0.15 and 0.65, and the maximum fluctuation segment is concentrated in the lower left area of ​​the image. This segment is identified as the segment with the strongest discreteness and recorded as the grayscale discrete amplitude of the contact area.

[0098] S413: Based on the grayscale discrete amplitude of the contact area, calculate the ratio to the path grayscale balance level, and calculate the grayscale distribution of the contact area using the formula:

[0099]

[0100] The contact uniformity coefficient CX of the crimping head is obtained to characterize the uniformity of the grayscale distribution of the wire path pixels in the contact area of ​​the crimping terminal. p Indicates the total number of pixels on the wire path, GX s Represents the grayscale data of the s-th pixel on the wire path, Indicates the grayscale balance level of all pixels on the wire path (i.e. the average grayscale value of the path pixels), VX g Indicates the grayscale discrete amplitude of the contact area, which is used to measure the overall amplitude of grayscale fluctuation in the area. p Indicates the continuity dispersion of the wire path pixels, which is used to describe the spatial distribution continuity characteristics of the path pixels;

[0101] First, define the number of path pixels as L p =5, the grayscale sequence of the path extracted from the image data is:

[0102] GX s ={120,130,125,135,140};

[0103] Calculate the balanced gray level of the sequence, that is, the gray mean Using weighted average method:

[0104]

[0105] Secondly, according to the difference between the grayscale of each pixel in the path and the balanced grayscale, the square difference of each item is calculated separately, that is, Get a series of square differences:

[0106] (120-130) 2 =100;

[0107] (130-130) 2 =0;

[0108] (125-130) 2 =25;

[0109] (135-130) 2 =25;

[0110] (140-130) 2 =100;

[0111] The total sum of squared differences is:

[0112] 100+0+25+25+100=250;

[0113] At the same time, let the grayscale discrete amplitude of the contact area VX g = 15, the continuity dispersion of the path pixels DX p =9, the dispersion is calculated by the continuous jump frequency statistics of the path pixels in the spatial distribution. If the jump between each two adjacent points in the path exceeds the set grayscale change threshold (such as 10), it is judged as a discontinuity, and the cumulative number is 2. Combined with the path length normalization, DX is obtained. p =9; according to the values ​​of each parameter, substitute into the following formula:

[0114]

[0115] The results show that under the current wire path, the concentration of its grayscale variation is still large under the comprehensive discrete benchmark, indicating that there is a certain discontinuous grayscale transition in the contact area of ​​the wire. If further optimization is needed, the continuity of the path layout or the grayscale adjustment within the region can be used to adjust the wire arrangement or the port design structure. The formula introduces the joint constraints of the squared term of the path grayscale deviation and the regional grayscale discrete factor and the path continuity factor to achieve quantitative modeling of the complex distribution of image grayscale information in the spatial path, thereby realizing visual evaluation and feature recognition of the crimping quality, thereby enhancing the entire system's ability to express the uniformity characteristics of the indenter image.

[0116] See also Figure 6 , the specific steps for obtaining the connection equilibrium risk indicator are:

[0117] S511: Based on the indenter contact uniformity coefficient, the grayscale uniformity in the indenter contact area image is analyzed. Combined with the contact area distribution and coverage consistency, the number of effective contact pixels within the wire path is calculated, and the pixel distribution within the area is determined to obtain the contact pixel density.

[0118] First, the contact area boundary is calibrated in the crimped terminal image, and a fixed image area is defined for pixel analysis. Then, the grayscale value data set of all pixels in the area is extracted, and the mean and standard deviation calculation operations are performed on the set to determine whether the grayscale value of the current area is concentrated in a certain grayscale interval. If more than 90% of the pixels in the area have grayscales between 100 and 160, then the grayscale distribution is determined to be uniform. On the contrary, if the pixel grayscale distribution span exceeds 100 grayscale levels, then the grayscale distribution is determined to be inconsistent. On this basis, the pixel coordinate set corresponding to the wire path is further extracted, and the pixel trajectory of each path is compared to obtain whether each pixel point on the path is located in the identified effective contact area range. If it is located in the area, then the pixel coordinate set is recorded. Pixels are valid contact points. After statistics are performed on all paths, the total number of valid contact pixels on the path is obtained. When continuing to perform regional distribution judgment, the coordinates of all valid contact pixels are projected into the contact area, and their spatial density is analyzed. If there are contact points in 5 consecutive rows of pixels within a certain range, and the number of pixels in each row is not less than 3 pixels, then the range is determined to be a pixel-dense distribution area. Combined with the above results, the concentration of pixel distribution in the indenter contact area is output for each path. In the actual image sample, the total length of a wire path is 180 pixels, of which there are 75 valid contact points, distributed in the middle right area of ​​the image. Their spatial distribution is concentrated in a 28×28 pixel square. After statistics, the contact pixel density of the path is obtained.

[0119] S512: Based on the contact pixel density, the ratio of the number of effective contact pixels to the extended area of ​​the wire path is compared with the extended area of ​​the wire path, and the distribution consistency of the pixels in the contact area is optimized to obtain a contact balanced distribution feature.

[0120] According to the number of effective contact pixels counted in the previous step, the entire coverage of the wire path in the crimping area is matched, the pixel coordinate set of the path in the image is extracted, and its spatial extension area is constructed. The boundary position is extracted and the total number of pixels covered is calculated as the extension area. The ratio of the effective contact pixel number to the path coverage area is then calculated. If the ratio is between 0.4 and 0.7, the contact is considered to be relatively balanced. If it is lower than 0.3, the contact is judged to be weak. If it is higher than 0.8, it is judged to be locally overcrowded. The distribution consistency analysis of the contact area is continued, and all effective contact points are divided into several rows and columns of grid blocks. The images in each block are counted. After calculating the number of pixels, the maximum and minimum difference is calculated. If the difference is less than 3, the contact distribution is considered consistent. Otherwise, it is considered uneven. Further pixel relocation operation is performed on the uneven area, its coordinates are recalibrated and the contact distribution map is drawn. Combined with the aforementioned contact uniformity analysis, the wire segment with a high effective contact area ratio and small distribution difference is selected as the optimal distribution feature segment. For example, in image number A001, the path area is 260 pixels and the number of contact pixels is 156, accounting for about 0.6. After grid statistics, the difference between the largest block and the smallest block is 2 pixels, which meets the consistency judgment. Finally, the path is identified as a good item with balanced contact distribution features.

[0121] S513: Based on the contact balance distribution characteristics, identify the wire harness endpoints that do not meet the balance requirements, determine the distribution status of the associated wire paths, and combine the wire harness number and distribution characteristics to obtain a connection balance risk index;

[0122] Traverse all the wire path contact conditions, filter out the path segments that do not meet the balance requirements, perform a joint judgment of area ratio and pixel consistency on each path, if the contact area ratio of a path is less than 0.3, and the contact pixels are mainly concentrated at the end of the path, and the regional difference exceeds 5 pixels, it is judged as not meeting the balance. Then, combined with the path identification information, extract the harness number corresponding to the path segment, and associate it with its position coordinates in the entire image, and count all harness endpoints that do not meet the balance requirements. If there are more than three abnormal endpoints in a single image, the corresponding sample of the image is marked For high-risk connection samples, we continue to extract the wire pixel trajectory of the endpoint, draw a grayscale extension map, and perform difference analysis on the boundary between the contact area and the non-contact area in each figure, record the location of the prominent change, and finally combine the harness number, distribution imbalance mark and the number of discrete coordinate points of the path to summarize the connection balance risk index of the current image. In a certain sample image, the harness path numbered C027 accounts for only 18%, and the effective pixels are concentrated in the area on the right side of the path. The number of contact boundary change points in the path map reaches 13. This path is marked as a high-risk item and participates in the generation of risk indicators.

[0123] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for detecting connector harness crimping defects based on image recognition, characterized in that: The following steps are involved: S1: Based on the image data of the connector harness crimping area, it is divided into multiple grids, the grayscale brightness mean of the edge and center pixels of each grid is compared to determine the local brightness distribution, and the grid brightness density gradient is integrated to obtain the local brightness distribution characteristics; S2: Based on the local brightness distribution characteristics, calculate its distribution in the crimping area of ​​the crimping terminal, analyze the normalized performance of the grid brightness density gradient, identify the spatial deviation aggregation area, determine its spatial correspondence, and obtain a three-dimensional abnormal aggregation factor; S3: Based on the three-dimensional abnormal aggregation factor, analyze the reflection characteristics of the image of the harness port area under the target light source, optimize the pixel reflection intensity sorting, screen the reflection intensity mutation area, compare the light energy density of the mutation area and the coating material area, and obtain the port exposure characteristic signal; S4: Based on the exposed port characteristic signal, determine the distribution of the wire path in the crimping terminal contact area image, analyze the wire pixel extension distribution, compare the grayscale uniformity of the contact area pixels, calculate the grayscale distribution of the contact area, and obtain the crimping terminal contact uniformity coefficient.

2. The method for detecting connector harness crimping defects based on image recognition according to claim 1, characterized in that: The local brightness distribution characteristics include brightness change characteristics, distribution uniformity, and structural clarity; the three-dimensional abnormal aggregation factor includes the number of abnormal areas, the degree of aggregation distribution, and the range of the connected area; the port exposure characteristic signal includes reflection intensity characteristics, metal exposure marks, and signal change types; the indenter contact uniformity coefficient includes contact area distribution, coverage consistency, and grayscale uniformity.

3. The method for detecting connector harness crimping defects based on image recognition according to claim 1, characterized in that: The steps for obtaining the local brightness distribution feature are specifically as follows: S111: Based on the image data of the crimping area of ​​the connector harness, the crimping area image is divided into a plurality of square grid areas, and a difference analysis is performed by comparing the grayscale brightness mean of each grid edge pixel with the center pixel to obtain a pixel brightness difference set; S112: Calculate the absolute value of the brightness mean difference of each grid based on the pixel brightness difference set, filter out grid areas that are greater than the brightness density standard, count the number of selected grids, and compare their distribution in the image space to obtain local brightness distribution characteristics.

4. The method for detecting connector harness crimping defects based on image recognition according to claim 1, characterized in that: The steps for obtaining the three-dimensional abnormal aggregation factor are specifically as follows: S211: Based on the local brightness distribution characteristics, calculate the difference between the edge and center brightness mean of each grid in the crimping area of ​​the crimping terminal, normalize the difference, and locate the brightness distribution of each grid in three-dimensional space to obtain grid brightness offset distribution data; S212: Based on the grid brightness offset distribution data, determine the continuous change of the normalized gradient of adjacent grids, identify the formed clustered areas, and perform connectivity detection to obtain spatial offset clustering performance; S213: According to the spatial offset aggregation performance, compare the aggregation center with the area corresponding to the boundary of the crimping area of ​​the crimping terminal. If the spatial position has an overlap, inclusion or adjacency relationship, mark the aggregation area as an abnormal aggregation area to obtain a three-dimensional abnormal aggregation factor.

5. The method for detecting connector harness crimping defects based on image recognition according to claim 1, wherein: The steps for obtaining the port exposure characteristic signal are specifically as follows: S311: Based on the three-dimensional abnormal aggregation factor, its regional distribution is analyzed, the order of pixel reflection intensity of the grid in the harness port area is optimized, the grayscale sequence change trend is calculated, and the sections with continuous drastic grayscale changes are screened to obtain a continuous mutation grayscale interval group; S312: comparing the average reflection intensity of the pixel set in the local space corresponding to the continuous mutation grayscale interval group, calculating the difference with the average reflection intensity of the coating material area, adjusting the proportional influence on the number of pixels, and obtaining the reflection intensity difference; S313: Analyze the pixel distribution in the reflection abnormality signal domain based on the reflection intensity difference, determine the spatial aggregation state, optimize the boundary consistency of the continuous abnormal area, compare the reflection characteristics with the metal area, identify the boundary features, and obtain the port exposure feature signal.

6. The method for detecting connector harness crimping defects based on image recognition according to claim 1, characterized in that: The steps for obtaining the indenter contact uniformity coefficient are specifically as follows: S411: Based on the exposed port characteristic signal, analyzing the pixel coordinate distribution of the wire path in the contact area image, calculating the grayscale data sequence of each pixel on the path, unifying the grayscale distribution range through normalization conversion, and comparing the spatial variation of grayscale between different pixels to obtain the wire grayscale extension distribution; S412: Calling the grayscale extension distribution of the wire, determining the distribution state of the standardized path grayscale sequence in the contact area, calculating the grayscale balance level in the area, analyzing the discrete relationship between the path grayscale and the balance level, optimizing the grayscale difference combination, identifying the optimal fluctuation amplitude, and obtaining the grayscale discrete amplitude of the contact area; S413: Based on the grayscale discrete amplitude of the contact area, calculate the ratio to the path grayscale balance level, calculate the grayscale distribution of the contact area, and obtain the indenter contact uniformity coefficient.

7. The method for detecting connector harness crimping defects based on image recognition according to claim 1, characterized in that: The steps also include: S5: Based on the indenter contact uniformity coefficient, calculate the correspondence with the wire extension length, analyze the ratio of the number of pixels in the contact area to the path extension area, determine whether the ratio is in a balanced range, identify the unbalanced endpoints, and obtain a connection balance risk index; The connection balance risk indicators include risk level, number of abnormal endpoints, and balance analysis results.

8. The method for detecting connector harness crimping defects based on image recognition according to claim 7, characterized in that: The steps for obtaining the connection balance risk indicator are specifically as follows: S511: Based on the indenter contact uniformity coefficient, the grayscale uniformity in the indenter contact area image is analyzed. In combination with the contact area distribution and coverage consistency, the number of effective contact pixels within the wire path is calculated, and the pixel distribution within the area is determined to obtain the contact pixel density. S512: Based on the contact pixel density, the ratio of the number of effective contact pixels to the extended area of ​​the wire path is compared with the extended area of ​​the wire path, and the distribution consistency of the pixels in the contact area is optimized to obtain a contact balanced distribution feature. S513: Based on the contact balance distribution characteristics, identify the wiring harness endpoints that do not meet the balance requirements, determine the distribution status of the associated wire paths, and obtain a connection balance risk index based on the wiring harness number and distribution characteristics.

Images