PCB welding spot defect detection system and method based on image recognition

Abstract

The invention relates to the technical field of PCB welding spot defect detection, and discloses a PCB welding spot defect detection system and method based on image recognition, and the system comprises an image preprocessing module which is used for obtaining an original image flow and dividing an interested detection area; the feature fusion module is used for extracting multi-modal features to form a fusion set; the defect judgment module is used for establishing a mapping index and obtaining a judgment result; the parameter calibration module is used for verifying the detection parameters and adjusting the mapping index; and the report output module is used for generating a defect detection report. The method comprises the steps of image preprocessing, feature fusion, defect discrimination, parameter calibration, report generation and the like. According to the system and the method, the PCB welding spot defects can be efficiently and accurately detected, the detection precision and stability are improved, a structured report is generated, and an effective solution is provided for PCB quality detection.

Description

Technical Field

[0001] The present invention relates to the technical field of PCB solder joint defect detection, and in particular to a PCB solder joint defect detection system and method based on image recognition. Background Art

[0002] In today's rapidly developing world of electronic information technology, PCBs (printed circuit boards), as core components of electronic devices, have a quality that directly impacts the performance and reliability of the entire device. Solder joints, the critical connections between components and the board, are crucial. Defects in solder joints, such as cold solder joints, insufficient solder, excessive solder, or bridging, can not only lead to poor circuit continuity and abnormal signal transmission, but in severe cases, can also cause failure of the entire device and even lead to serious safety incidents.

[0003] As electronic components continue to become increasingly miniaturized and integrated, the number of solder joints on PCBs is increasing while their size is shrinking. This places higher demands on solder joint defect detection. Traditional manual visual inspection methods are not only inefficient and unable to meet the demands of large-scale production, but are also significantly affected by the subjective factors of the inspectors, resulting in unstable detection accuracy and prone to missed detections and false detections.

[0004] While existing automated inspection technologies, such as X-ray and ultrasonic testing, have improved inspection efficiency and accuracy to a certain extent, these technologies often suffer from high equipment costs, slow inspection speeds, and poor detection performance for certain defect types. For example, X-ray inspection equipment is expensive and lacks optimal detection performance for some surface defects, while ultrasonic testing requires a high inspection environment and is relatively complex to operate.

[0005] The application of image recognition technology in industrial inspection has provided new insights into PCB solder joint defect detection. However, existing image recognition-based PCB solder joint defect detection technologies still have numerous shortcomings. Firstly, during the image preprocessing stage, it is difficult to accurately delineate the inspection area of ​​interest, and this technology is susceptible to interference from background noise and other components, resulting in inaccurate feature extraction. Secondly, during the feature extraction process, only a single feature type, such as geometric features or grayscale features, is often considered, which fails to fully reflect the true state of the solder joint, resulting in low accuracy and reliability of defect identification results.

[0006] Existing inspection systems lack effective parameter calibration mechanisms, making it difficult to ensure the stability of inspection results when the inspection environment or process parameters change. Furthermore, inspection report generation is not intelligent enough to properly sort and analyze defects based on their severity, hindering subsequent defect remediation and quality improvement.

[0007] Therefore, improving the efficiency, accuracy, and stability of PCB solder joint defect detection and achieving rapid and accurate identification and classification of solder joint defects has become a pressing issue in the current PCB manufacturing industry. This invention aims to provide a PCB solder joint defect detection system and method based on image recognition to address these issues in the prior art. Summary of the Invention

[0008] The purpose of the present invention is to provide a PCB solder joint defect detection system and method based on image recognition to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a PCB solder joint defect detection system based on image recognition, the system comprising: The image preprocessing module is used to obtain the original image stream of the solder joints to be inspected on the PCB board and divide the inspection area of ​​interest of the solder joints to be inspected according to the process parameters of the PCB board; The feature fusion module is used to extract multimodal features from the original image stream, obtain the geometric features, grayscale distribution features, and texture gradient features of the solder joints to be detected, and form a solder joint feature fusion set; The defect discrimination module is used to obtain the discrimination benchmark of the solder joint feature fusion set, establish the mapping index between the feature fusion set and the defect type, calculate the matching coefficient between the solder joint feature fusion set and the standard sample, and obtain the defect type discrimination result; The parameter calibration module is used to verify the consistency of the inspection parameters of the solder joint to be inspected based on the defect type identification results, identify abnormal offsets in the inspection parameters, and adjust the mapping index between the feature fusion set and the defect type based on the abnormal offsets; The report output module is used to extract the defect type and spatial coordinates of the solder joint to be inspected based on the adjusted feature fusion set and the mapping index of the defect type, and generate a defect inspection report according to the structured template.

[0010] Preferably, the implementation of the image preprocessing module includes: For any solder joint to be inspected on the PCB board, an industrial line scan camera matching the process parameters is called; The original image stream of the solder joint is collected by an industrial line scan camera, and the image contrast is enhanced by using histogram equalization operation; The edge gradient value of the solder joint area in the enhanced image is identified, and the continuous pixel group with gradient value greater than the threshold is extracted as the detection area of ​​interest of the solder joint to be detected.

[0011] Preferably, the implementation method of extracting a continuous pixel group whose gradient value is greater than a threshold value further includes: Convert the enhanced image into HSL color space and calculate the brightness component of each pixel; Extract pixel points whose brightness components are within a preset range as candidate edge points, and connect adjacent candidate edge points to form a closed area; Taking the closed area as the boundary, the pixel matrix of the corresponding area in the HSL image is intercepted as the detection area of ​​interest of the solder joint to be detected.

[0012] Preferably, the implementation method of dividing the inspection area of ​​interest of the solder joint to be inspected also includes: Mark the coordinates of the detection area of ​​interest, analyze the row and column coordinates of the area on the PCB board, perform spatial projection of the detection area of ​​interest according to the row and column coordinates, and establish a positioning association relationship between the detection area of ​​interest and the pad position on the PCB board.

[0013] Preferably, the feature fusion module is implemented as follows: Invoking a geometric morphology algorithm, a grayscale statistics algorithm, and a texture analysis algorithm for the solder joint to be inspected to generate multiple unfused feature extraction results, where the unfused feature extraction results represent geometric parameters, grayscale parameters, and texture parameters that are not associated with the defect type; It is determined whether multiple unfused feature extraction results are valid feature subsets. If they are valid feature subsets, the valid feature subsets are merged into a solder joint feature fusion set.

[0014] Preferably, the implementation method of establishing a mapping index between a feature fusion set and a defect type includes: using the geometric feature description, grayscale distribution value and texture gradient pattern in the weld feature fusion set, combined with the definition parameters of the defect type and the identification code of the defect type, to establish a mapping index between the feature fusion set and the defect type.

[0015] Preferably, the defect identification module is implemented as follows: Cluster the solder joint feature fusion set and defect types according to defect shape, defect size, and impact level, and set the main cluster center after clustering as the feature discrimination benchmark; Extracting the core feature parameters of the feature discrimination benchmark, calculating the parameter similarity between the core feature parameters, and setting a feature sequence related to the parameter similarity between the core feature parameters; Using the feature sequence related to the parameter similarity between each core feature parameter, the defect features in the sequence are extracted, and the longest common feature subsequence between each defect feature is set, and the matching value of the longest common feature subsequence is set as the defect feature matching coefficient; The defect feature matching coefficients between the defect features are distributed according to the occurrence frequency of each defect feature to set the defect type discrimination result.

[0016] Preferably, the parameter calibration module is implemented as follows: Extracting the frequency distribution of each defect feature from the defect type discrimination result; setting a calibration path for the defect type discrimination result according to the detection period corresponding to the frequency distribution of each defect feature; Perform curve fitting on the calibration path of each defect feature in the defect type discrimination result to obtain a fitted calibration path, and set the probability value of the fitted path in each detection period as the discrimination confidence of the defect type discrimination result; The confidence level of the defect type identification result is compared with the standard reference value of the detection parameter, the existing offset difference is marked, and the detection parameter is corrected according to the offset difference to complete the calibration of the detection parameter.

[0017] Preferably, the report output module is implemented as follows: Based on the adjusted feature fusion set and the mapping index of the defect type, the defect type of the solder joint to be inspected is extracted, and the defect type is prioritized according to the severity level of the defect type to obtain the defect processing order of the solder joint to be inspected; the defect type, spatial coordinates and defect processing order are integrated in a tabular form to generate a defect detection report.

[0018] Preferably, the present invention further includes a PCB solder joint defect detection method based on image recognition, which is applied to the above-mentioned PCB solder joint defect detection system based on image recognition, and the method includes the following steps: Obtain the original image stream of the solder joints to be inspected on the PCB board, and divide the inspection area of ​​interest of the solder joints to be inspected according to the process parameters of the PCB board; Perform multimodal feature extraction on the original image stream to obtain the geometric features, grayscale distribution features, and texture gradient features of the solder joints to be detected, forming a solder joint feature fusion set; Obtain the discrimination benchmark of the solder joint feature fusion set, establish a mapping index between the feature fusion set and the defect type, calculate the matching coefficient between the solder joint feature fusion set and the standard sample, and obtain the defect type discrimination result; Using the defect type identification results as a reference, the consistency of the inspection parameters of the solder joint to be inspected is verified, abnormal offsets in the inspection parameters are identified, and the mapping index between the feature fusion set and the defect type is adjusted based on the abnormal offsets. Based on the adjusted mapping index, the defect type and spatial coordinates of the solder joint to be inspected are extracted, and a defect inspection report is generated according to the structured template.

[0019] Compared with the prior art, the present invention has the following beneficial effects: During image preprocessing, an industrial line scan camera matched to the process parameters is used to capture the raw image stream. Histogram equalization is then applied to enhance image contrast. Combined with HSL color space analysis and edge gradient value extraction, the region of interest (ROI) for the solder joints to be inspected can be precisely delineated. This approach effectively eliminates background noise and interference from other components, laying a solid foundation for subsequent feature extraction and significantly improving the accuracy and reliability of image preprocessing.

[0020] The feature fusion module uses geometric morphology, grayscale statistics, and texture analysis algorithms to perform multimodal feature extraction on the original image stream, acquiring geometric features, grayscale distribution features, and texture gradient features to form a fused set. This multi-feature fusion approach comprehensively and accurately reflects the true state of the solder joint, overcoming the limitations of single-feature extraction in existing technologies. The extracted features are richer and more comprehensive, providing a more comprehensive basis for defect identification.

[0021] The defect identification module clusters the solder joint feature fusion set and defect type, sets a feature identification benchmark, calculates the similarity between core feature parameters, extracts defect features, and sets the matching value of the longest common feature subsequence as the defect feature matching coefficient, thereby accurately identifying the defect type. This process fully considers factors such as defect shape, size, and impact level, improving the accuracy and reliability of defect identification and enabling more precise identification of various types of solder joint defects.

[0022] The parameter calibration module uses defect type identification results as a reference to verify the consistency of inspection parameters, identify abnormal offsets, and adjust mapping indices. This mechanism enables the inspection system to automatically calibrate inspection parameters based on actual inspection results, adapting to changes in the inspection environment and process parameters. This improves the inspection system's stability and adaptability, ensuring high inspection accuracy under varying conditions.

[0023] The report output module extracts defect types and spatial coordinates based on the adjusted mapping index, prioritizes defects by severity, and generates a structured defect detection report in tabular format. This approach not only improves the readability and practicality of the report but also provides a scientific basis for subsequent defect remediation and quality improvement, helping to improve production efficiency and product quality.

[0024] Through the coordinated work of various modules, the present invention realizes efficient and accurate detection of PCB solder joint defects, improves detection efficiency and accuracy, enhances the stability and adaptability of the detection system, provides reliable quality assurance for PCB production and manufacturing, and has significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1This is a working principle diagram of the PCB solder joint defect detection system based on image recognition according to the present invention; Figure 2 This is the working principle diagram of the image preprocessing module; Figure 3 The working principle diagram for extracting continuous pixel groups with gradient values ​​greater than a threshold; Figure 4 This is the working principle diagram of the defect identification module. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0027] See also Figure 1-Figure 4 The present invention relates to a PCB solder joint defect detection system based on image recognition, which includes an image preprocessing module, a feature fusion module, a defect discrimination module, a parameter calibration module, and a report output module. The specific implementation is as follows: The image preprocessing module is used to obtain the original image stream of the solder joints to be inspected on the PCB board and divide the inspection area of ​​interest of the solder joints to be inspected according to the process parameters of the PCB board. Specifically, for any solder joint to be inspected on the PCB board, an industrial linear array camera that matches the process parameters is called to collect the original image stream of the solder joint through the camera. Then, a histogram equalization operation is used to enhance the image contrast. After that, the edge gradient value of the solder joint area in the enhanced image is identified, and a continuous pixel group with a gradient value greater than a threshold is extracted as the inspection area of ​​interest for the solder joint to be inspected. At the same time, the coordinates of the inspection area of ​​interest are also marked, and the row and column coordinates of the area on the PCB board are analyzed. The inspection area of ​​interest is spatially projected according to the row and column coordinates, thereby establishing a positioning association relationship between the inspection area of ​​interest and the position of the PCB pad.

[0028] The feature fusion module performs multimodal feature extraction on the raw image stream, acquiring the geometric features, grayscale distribution features, and texture gradient features of the solder joints to be inspected, thereby forming a fused solder joint feature set. This is achieved by invoking the geometric morphology algorithm, grayscale statistics algorithm, and texture analysis algorithm for the solder joints to be inspected, generating multiple unfused feature extraction results. These unfused feature extraction results represent geometric, grayscale, and texture parameters not associated with the defect type. The module then determines whether these unfused feature extraction results constitute valid feature subsets. If so, these valid feature subsets are merged into the fused solder joint feature set.

[0029] The defect discrimination module is used to obtain the discrimination benchmark of the solder joint feature fusion set, establish a mapping index between the feature fusion set and the defect type, calculate the matching coefficient between the solder joint feature fusion set and the standard sample, and obtain the defect type discrimination result. Specifically, the solder joint feature fusion set and the defect type are clustered according to the defect morphology, defect size, and impact level. The main cluster center after clustering is set as the feature discrimination benchmark, the core feature parameters of the feature discrimination benchmark are extracted, and the parameter similarity between each core feature parameter is calculated. The feature sequence related to the parameter similarity between each core feature parameter is set. The feature sequence is used to extract the defect features in the sequence, and the longest common feature subsequence between each defect feature is set. The matching value of the longest common feature subsequence is set as the defect feature matching coefficient. Finally, the defect feature matching coefficient between each defect feature is set according to the frequency distribution of each defect feature to determine the defect type discrimination result.

[0030] The parameter calibration module is used to verify the consistency of the inspection parameters of the solder joints to be inspected, identify abnormal offsets in the inspection parameters, and adjust the mapping index between the feature fusion set and the defect type based on the abnormal offsets, with the defect type discrimination results as the reference. The specific operation is to extract the frequency distribution of each defect feature from the defect type discrimination results, set the calibration path of the defect type discrimination results according to the detection period corresponding to the frequency distribution of each defect feature, perform curve fitting on the calibration path of each defect feature in the defect type discrimination results, obtain the fitted calibration path, set the probability value of the fitting path in each detection period as the discrimination confidence of the defect type discrimination result, compare the discrimination confidence with the standard reference value of the detection parameter, mark the existing offset difference, and correct the detection parameters according to the offset difference to complete the calibration of the detection parameters.

[0031] The report output module is used to extract the defect type and spatial coordinates of the solder joints to be inspected based on the adjusted feature fusion set and the mapping index of the defect type, and then generate a defect inspection report according to a structured template. Specifically, the defect type of the solder joint to be inspected is extracted based on the adjusted mapping index, and the defect types are prioritized by their severity level to determine the defect handling sequence for the solder joints to be inspected. The defect type, spatial coordinates, and defect handling sequence are then integrated in a tabular format to generate a defect inspection report.

[0032] Example 1: In the image preprocessing module, for any solder joint to be inspected on a PCB, an industrial line scan camera matching the PCB's process parameters must be first used. The selection of an industrial line scan camera must strictly adhere to the specific process requirements of the PCB. For example, when the solder joints on a PCB are small and dense, an industrial line scan camera with high resolution and high frame rate should be selected to ensure clear capture of the original image stream of the solder joints. Different models of industrial line scan cameras vary in terms of sensor size, pixel resolution, and scanning frequency. Only by matching the process parameters can the captured images meet the accuracy requirements of subsequent inspections.

[0033] The solder joints are imaged using a selected industrial line scan camera to generate a raw image stream. In actual production environments, factors such as lighting conditions and the relative position of the camera and PCB board can result in low contrast in the raw image stream, making the distinction between the solder joints and the background unclear. Therefore, histogram equalization is performed on the raw image stream to enhance image contrast. Histogram equalization works by readjusting the grayscale distribution of the image, expanding pixel values ​​originally concentrated in a specific grayscale range to a wider range. This allows for details in dark areas of the image to be revealed, and for bright areas to have richer gradations. This improves the contrast between the solder joint area and the background, laying the foundation for subsequent edge detection and region extraction.

[0034] After image contrast enhancement, it is necessary to identify the edge gradient values ​​of the solder joint areas in the enhanced image. Edge gradient values ​​reflect the severity of pixel grayscale changes within the image. Solder joint edges typically have larger gradient values. Classic gradient operators, such as the Sobel operator or the Prewitt operator, can be used for this purpose. For example, the Sobel operator performs convolution operations on the image horizontally and vertically using two 3×3 convolution kernels, respectively, to obtain horizontal and vertical gradients. The gradient value of each pixel is then determined by combining these two values. After calculating the gradient values ​​for each pixel, an appropriate threshold is set to extract consecutive pixels with gradient values ​​greater than the threshold. The threshold setting requires comprehensive consideration of factors such as the PCB material, solder joint shape, and production environment. A reasonable range can typically be determined by analyzing a large number of standard solder joint images.

[0035] In order to further improve the accuracy of extracting the detection area of ​​interest, the enhanced image can also be converted to the HSL color space. The HSL color space represents color as three components: hue (H), saturation (S), and brightness (L). Compared with the traditional RGB color space, it is more in line with the human perception of color. In the HSL color space, the brightness component of each pixel is calculated. Since solder joints usually have specific brightness characteristics, pixels with brightness components within a preset range can be extracted as candidate edge points. The determination of the preset range requires reference to the brightness distribution characteristics of standard solder joints in the HSL color space. For example, by performing statistical analysis on the brightness components of multiple standard solder joint images, a brightness range that can contain the vast majority of solder joint pixels is determined.

[0036] After extracting candidate edge points, adjacent candidate edge points need to be connected to form a closed region. During this connection process, the spatial relationship between pixels and the consistency of gradient direction must be considered to ensure that the resulting closed region accurately encompasses the solder joint. Using this closed region as the boundary, the pixel matrix corresponding to the HSL image is captured. This pixel matrix represents the region of interest (ROI) for the solder joint to be inspected. Processing in the HSL color space allows for better utilization of the solder joint's color characteristics, improving the accuracy of ROI extraction and avoiding misdetection due to factors such as uneven lighting.

[0037] After acquiring the inspection area of ​​interest, its coordinates need to be annotated. Specifically, the row and column coordinates of the area on the PCB are analyzed. This can be achieved by converting between the PCB coordinate system and the image coordinate system. Typically, during installation of the inspection system, the camera position and the PCB coordinate system are calibrated to establish a mapping relationship between the two. The inspection area of ​​interest is spatially projected based on its row and column coordinates. This spatial projection determines the specific location of the area on the PCB.

[0038] Establishing a positioning association between the inspection area of ​​interest and the pad location on the PCB is a crucial step. Through the aforementioned coordinate annotation and spatial projection, the inspection area of ​​interest in the image can be accurately mapped to the actual pad location on the PCB. This positioning association ensures that subsequent inspection results for solder joint defects correspond to their specific locations on the PCB, facilitating the precise location and repair of defective solder joints during production. For example, if a defect is detected in a particular inspection area of ​​interest, the positioning association can be used to quickly locate the defect's specific location on the PCB, improving production efficiency and quality control accuracy.

[0039] Throughout the image preprocessing process, each step must strictly adhere to the aforementioned procedures to ensure that the captured inspection area of ​​interest accurately reflects the actual weld condition. The correct selection of an industrial line scan camera ensures image acquisition quality, histogram equalization enhances image readability, edge gradient identification and candidate edge point extraction ensure the accuracy of the region of interest, and coordinate annotation and positioning associations provide the foundation for subsequent defect detection and location. These various steps work together to form a complete implementation of the image preprocessing module, laying a solid foundation for subsequent operations such as feature fusion and defect identification.

[0040] Example 2: The implementation method of the feature fusion module needs to be expanded from the underlying logic of multimodal feature extraction. After obtaining the original image stream processed by the image preprocessing module, the system needs to synchronously call the geometric morphology algorithm, grayscale statistics algorithm and texture analysis algorithm for the solder joint to be detected. Among them, the geometric morphology algorithm is based on the theory of mathematical morphology. By designing structural elements of different shapes (such as circles, squares, diamonds, etc.), the solder joint image is corroded, expanded, opened, closed and other operations are performed to extract the geometric feature parameters of the solder joint. For example, by calculating the area, perimeter, equivalent diameter, circularity (i.e. 4π×area / perimeter) of the solder joint area 2 ), rectangularity (area / minimum circumscribed rectangle area) and other parameters to describe the shape characteristics of the solder joint; by identifying the inflection points, sharp corners and other feature points on the edge of the solder joint, the contour morphology of the solder joint is reflected.

[0041] Grayscale statistics algorithms are primarily used to analyze the grayscale distribution characteristics of solder joint images. The algorithm first performs a statistical analysis of the grayscale values ​​of the solder joint image, calculating basic statistical quantities such as the grayscale mean, variance, skewness, and kurtosis. The grayscale mean reflects the overall brightness of the solder joint, the variance reflects the dispersion of the grayscale values, the skewness describes the symmetry of the grayscale distribution, and the kurtosis indicates the steepness of the grayscale distribution. Furthermore, a grayscale histogram can be generated. By analyzing the histogram's shape, peak position, and distribution range, the grayscale characteristics of the solder joint can be further analyzed. For example, if a solder joint has a cold solder joint defect, its grayscale distribution may exhibit a peak shift or distribution range change that differs from that of a normal solder joint.

[0042] Texture analysis algorithms are used to extract texture gradient features from solder joint images. These algorithms analyze grayscale variation patterns in local regions of the image. Common texture analysis methods include gray-level co-occurrence matrices (GLCMs), local binary patterns (LBPs), and Gabor filters. Taking the GLCM as an example, this method constructs a co-occurrence matrix by calculating the frequency of grayscale combinations of two pixels in an image at specific directions and distances. This matrix then extracts texture feature parameters such as energy, entropy, contrast, and correlation. These parameters can reflect characteristics such as the surface roughness and texture directionality of the solder joint. For example, the texture of a normal solder joint typically exhibits a certain regularity and uniformity, while a solder joint with defects such as insufficient solder may exhibit a chaotic texture, with the corresponding texture feature parameters significantly altered.

[0043] Through the parallel operation of the three algorithms described above, the system generates multiple unfused feature extraction results. These results exist in the form of numerical values ​​or vectors, corresponding to geometric parameters, grayscale parameters, and texture parameters that are not associated with the defect type. For example, the geometric morphology algorithm may output geometric parameters such as the solder joint area of ​​12.5 square pixels, the perimeter of 15.3 pixels, and the circularity of 0.82; the grayscale statistics algorithm may obtain grayscale parameters such as the grayscale mean of 120 and the variance of 35.6; and the texture analysis algorithm may extract texture parameters such as energy of 0.75 and contrast of 23.4. These unfused feature parameters describe the image characteristics of the solder joint from different dimensions, but have not yet formed a comprehensive feature representation.

[0044] Next, the system needs to judge the validity of these unfused feature extraction results to determine whether they constitute a valid feature subset. The basis for the validity judgment mainly includes the discriminability of the features and the correlation between features. The discriminability of a feature refers to the ability of the feature to distinguish different types of defects, that is, whether there are significant differences in the values ​​of different defect types on the feature. For example, for two defects, cold solder joints and excessive solder, if the values ​​of a certain geometric feature parameter are significantly different between the two, then the feature has a higher degree of discrimination. The correlation between features refers to the degree of association between different features. If there is a high correlation between two features (such as a correlation coefficient greater than 0.8), it means that they may contain similar information. When fusing features, it can be considered to retain one of them to avoid information redundancy.

[0045] The specific implementation process for determining validity is as follows: First, for each unfused feature, a large amount of solder joint sample data of known defect types is collected, and the value of each sample for that feature is calculated. Next, statistical methods (such as t-tests and analysis of variance) are used to analyze whether the mean differences between different defect types for that feature are statistically significant to assess the feature's discriminatory power. Correlation coefficients (such as the Pearson correlation coefficient) are calculated between features to determine their inter-feature relevance. Based on preset discrimination and correlation thresholds, the validity of the feature is determined. For example, a discrimination p-value less than 0.05 is considered valid, and a correlation absolute value less than 0.7 is considered acceptable.

[0046] When it is determined that multiple unfused feature extraction results constitute a valid feature subset, the system will perform a feature fusion operation to merge these valid feature subsets into a weld feature fusion set. Feature fusion uses a splicing method, which connects all valid feature parameters in a certain order into a high-dimensional feature vector. For example, if the valid features include 3 geometric parameters, 4 grayscale parameters, and 5 texture parameters, the dimension of the fused feature vector is 3+4+5=12 dimensions. During the splicing process, attention should be paid to the dimensionality unification of feature parameters. For features with different dimensions, they can be first standardized (such as Z-score standardization or Min-Max standardization) to convert them into dimensionless values ​​to avoid the impact of dimensional differences on subsequent defect identification.

[0047] The construction of a feature fusion set is of great significance. It integrates feature information originally scattered across different modalities into a comprehensive feature representation, enabling a more comprehensive and accurate description of the image characteristics of solder joints. For example, geometric features reflect the shape and outline of the solder joint, grayscale features reveal the distribution of light and dark in the solder joint, and texture features reveal the microstructure of the solder joint surface. The fusion of these three allows the system to analyze solder joints from multiple dimensions, improving the accuracy and reliability of defect detection. Furthermore, feature fusion and validity screening can reduce the dimensionality of features, lowering the computational complexity of the subsequent defect identification module and improving the system's operational efficiency.

[0048] Throughout the implementation of the feature fusion module, parameter settings and algorithm selection at each stage require adjustment and optimization based on the specific PCB type, solder joint process requirements, and defect type. For example, the grayscale characteristics of solder joints may differ for PCBs made of different materials, requiring corresponding adjustment of the grayscale statistics algorithm parameters. For solder joints of varying shapes, the structural element selection of the geometric morphology algorithm also requires optimization. Only through reasonable algorithm configuration and a rigorous feature screening and fusion process can the generated solder joint feature fusion set accurately reflect the actual condition of the solder joints and provide reliable feature input for subsequent defect identification.

[0049] Example 3: The implementation of the defect identification module begins with establishing a discrimination benchmark. The system then clusters the acquired solder joint feature fusion set with known defect type data according to three dimensions: defect morphology, defect size, and impact level. Defect morphology encompasses the contour features of the solder joint, such as the presence of cracks, notches, and burrs; defect size includes quantitative parameters such as area, length, and depth; and impact level is determined by the degree to which the defect affects the electrical performance and mechanical strength of the PCB board, such as fatal defects, severe defects, and general defects. Clustering uses an iteratively optimized clustering algorithm. Initially, several cluster centers are set based on prior knowledge. Each feature fusion set is then assigned to the nearest category by calculating its distance from the cluster center. The category center is then recalculated, and this process is repeated until the cluster center stabilizes or the preset number of iterations is reached.

[0050] After clustering is complete, the primary cluster center within each cluster is set as the feature discrimination benchmark. The primary cluster center is determined based on the sample density within that cluster. Typically, the center of the region with the highest sample density is selected as the primary cluster center to ensure it represents the typical characteristics of that category. For example, in a cluster of cold solder joints, the primary cluster center should include a typical combination of geometric features (such as small solder joint area and low circularity), grayscale features (high grayscale mean), and texture features (high texture entropy).

[0051] Extract the core feature parameters for the feature discrimination benchmark. The selection of core feature parameters must adhere to the principles of effectiveness and representativeness. By analyzing the contribution of each feature parameter to defect classification, feature importance assessment methods (such as information gain and the Gini index) are used to select parameters that are highly discriminative of defect types. For example, when distinguishing bridging defects from normal solder joints, core feature parameters such as solder joint spacing (in geometric features), grayscale differences between adjacent solder joints (in grayscale features), and texture correlation between solder joints (in texture features) may serve as core feature parameters.

[0052] Parameter similarity is calculated between each core feature parameter. This calculation method varies depending on the feature type: for numerical features, Euclidean distance or cosine similarity is used; for categorical features, the chi-square test or information entropy is used. This similarity calculation constructs a correlation matrix between the core feature parameters, reflecting the dependencies and synergies between them. For example, a high similarity between solder joint area and grayscale mean indicates a strong correlation between these two feature parameters in describing the solder joint status.

[0053] Based on parameter similarity, a feature sequence associated with each core feature parameter is established. The construction of the feature sequence uses the strength of the correlation between parameters as a weight. Feature parameters with high similarity are arranged in a certain order to form a sequence structure that reflects the inherent connections between defect characteristics. For example, when describing an insufficient solder defect, the feature sequence might be solder joint area, grayscale mean, and texture contrast. These three feature parameters are sorted by the strength of their correlation with the defect, reflecting the typical evolution order of the defect's characteristics in terms of geometry, grayscale, and texture.

[0054] The constructed feature sequence is used to extract defect features within the sequence. Using a sliding window or sequence pattern matching approach, subsequences matching the known defect feature pattern are identified within the feature sequence. For example, the feature pattern for an excess solder defect is assumed to be "area exceeding a threshold → grayscale mean below a threshold → texture energy exceeding a threshold." When a subsequence matching this pattern appears within the feature sequence, the corresponding defect feature is extracted.

[0055] The longest common feature subsequence between each defect signature is determined. This is determined using a dynamic programming algorithm. The length of the common subsequence between feature sequences corresponding to different defect signatures is calculated to find the longest common portion. The matching value of this longest common feature subsequence is set as the defect signature matching coefficient. The matching coefficient ranges from [0, 1]. A higher value indicates a higher degree of match between the features of the solder joint to be inspected and the standard defect signature. For example, if the longest common subsequence length between the feature sequence of the solder joint to be inspected and the standard feature sequence of a certain defect type is 8, and the standard sequence length is 10, the matching coefficient is 0.8.

[0056] The defect type discrimination result is determined by assigning matching coefficients between each defect feature based on the frequency of occurrence of each defect feature. First, the frequency of occurrence of each defect feature in the historical inspection data is calculated. Defect features with higher frequencies are given higher weights in the discrimination result. For example, if a defect feature has a 30% frequency in the historical data, its corresponding matching coefficient will have a weight of 0.3 in the final discrimination result. A weighted summation is used to calculate the comprehensive matching score between the inspected solder joint and each defect type. The defect type with the highest score is the discrimination result.

[0057] During implementation, the initial parameters of the clustering algorithm, the screening thresholds for core feature parameters, and the selection of similarity calculation methods all need to be adjusted based on the specific PCB production process and defect type. For example, for high-density PCBs, where solder joint spacing is smaller, the selection of core feature parameters requires greater attention to the geometric relationships between solder joints. For high-frequency circuit boards, defect impact assessments should prioritize electrical performance indicators. Furthermore, the system must regularly update cluster centers and feature discrimination benchmarks based on new inspection data to adapt to changes in production processes and emerging defect types, ensuring the accuracy and timeliness of defect identification results.

[0058] The entire defect identification process constructs a discrimination benchmark through multi-dimensional clustering, combining similarity analysis and sequence matching of feature parameters to achieve a mapping from feature fusion sets to defect types. This process not only considers the discriminative role of individual feature parameters but also fully leverages the correlations between feature parameters and sequence characteristics. This allows for a more comprehensive capture of the complex characteristics of solder joint defects, improving the robustness and accuracy of defect identification. Furthermore, by incorporating the frequency distribution of defect features as weights, the identification results are more consistent with actual defect occurrence patterns in production, enhancing the system's practicality.

[0059] Example 4: The implementation of the parameter calibration module is based on the defect type identification results, from which the frequency distribution of each defect feature is extracted. For example, in the inspection of a batch of PCB boards, the identification results show that the defect feature of cold solder joints appeared 20 times, the defect feature of insufficient solder appeared 15 times, the defect feature of bridging appeared 8 times, and the defect feature of other types appeared 12 times. The system will record the number of occurrences of these different defect features in chronological order of inspection, forming the corresponding frequency distribution data. The defect features here can be specific characteristic parameters such as solder joint area less than the standard value, abnormal grayscale mean, and disordered texture gradient pattern. The frequency distribution of each feature can reflect the probability of occurrence of that feature within different inspection time periods.

[0060] The defect type discrimination results are divided into detection periods corresponding to the frequency distribution of each defect characteristic, and the calibration path for the defect type discrimination results is set. The detection period can be divided into time units such as hours, shifts, and dates. For example, one detection period is per hour, and the frequency of occurrence of each defect characteristic within each period is counted. Suppose that during the detection period from 8:00 to 9:00 in the morning, the cold solder defect characteristic appears 5 times, and the insufficient solder defect characteristic appears 3 times; during the period from 9:00 to 10:00, the cold solder characteristic appears 4 times, the insufficient solder characteristic appears 4 times, and the bridging characteristic appears 2 times. Based on the frequency distribution within these time periods, the system will construct a time-varying frequency curve for each defect characteristic. This curve forms the basis of the calibration path.

[0061] Perform curve fitting on the calibration path of each defect feature in the defect type identification results. When performing curve fitting, a function model suitable for changes in frequency distribution is used, such as a polynomial function, exponential function, or Gaussian function. For example, if the frequency of occurrence of a defect feature shows a trend of first increasing and then decreasing during the detection period, a quadratic polynomial function can be used for fitting; if it shows a gradually increasing trend, an exponential function can be used for fitting. During the fitting process, the function parameters are adjusted using methods such as the least squares method to make the fitting curve as close as possible to the actual frequency distribution data points. The fitted calibration path can smoothly represent the changing trend of the defect feature frequency over time, eliminate fluctuations caused by accidental factors, and thus more clearly reflect the underlying laws in the detection process.

[0062] The probability value of the fitted path in each inspection period is set as the confidence level of the defect type determination result. For example, after curve fitting, during the inspection period from 10:00 AM to 11:00 AM, the fitted path probability value for the cold solder defect feature is 0.65, the probability value for the insufficient solder defect feature is 0.30, and the probability values ​​for other features are 0.05. These probability values ​​represent the confidence level in the determination result for the corresponding defect type during that period. The determination confidence level reflects the system's assessment of the reliability of the current inspection result. The higher the probability value, the more reliable the defect type determination.

[0063] The confidence level of the defect type determination result is compared with the standard reference values ​​of the inspection parameters. These are pre-set based on the PCB production process requirements. For example, the standard exposure time for industrial line scan cameras is 10ms, the standard focal length is 50mm, the standard edge gradient threshold in the feature extraction algorithm is 80, and the standard grayscale distribution preset range is a brightness component between 60-120. For example, if the confidence level for detecting a cold solder joint suddenly increases during a certain inspection period, while the actual camera exposure time is 15ms, a deviation of 5ms from the standard value of 10ms, the system will flag this deviation.

[0064] Correct the detection parameters based on the offset difference to complete calibration. During correction, adjust the parameter value according to the direction and magnitude of the offset. For example, if the exposure time offset is 5ms, adjust it back to around 10ms. Set a reasonable adjustment step size based on the actual situation to avoid over-adjustment and new deviations. If multiple detection parameters deviate simultaneously, analyze the correlation between them and correct them in order of priority. For example, if both the camera focal length and exposure time deviate from the standard value, and if the focal length offset has a more critical impact on image quality, calibrate the focal length parameter first.

[0065] In a specific implementation, suppose that during the inspection period from 2:00 to 3:00 p.m. on a certain PCB production line, the system identifies a large number of solder joints as having insufficient solder defects, with a confidence level of 0.85. At this time, the inspection parameters for this period are checked, and it is found that the preset range of the brightness component of the grayscale statistics algorithm in the feature extraction algorithm is 40-80, while the standard reference value is 60-120. Because the lower limit of the preset range is lower than the standard value, the system may mistakenly identify the brightness component of normal solder joints as abnormal, thereby overestimating the frequency of insufficient solder defects. The system marks the offset difference as a lower limit offset of 20 and an upper limit offset of 40, and then adjusts the preset range of the brightness component to 60-120 to make the inspection parameters consistent with the standard reference value.

[0066] After calibration, the system recalculates the confidence level for that time period and observes whether the frequency of defect signatures returns to normal during subsequent inspection periods. For example, after adjusting the preset range for the brightness component, the confidence level for insufficient solder defects dropped to 0.20 during the inspection period from 3:00 PM to 4:00 PM, approaching normal levels and indicating that the calibration of the inspection parameters was effective. If the frequency of defect signatures does not improve after calibration, further investigation is necessary to determine whether other inspection parameters have shifted or whether new influencing factors, such as camera lens contamination or PCB conveyor vibration, are present. This will allow for more comprehensive parameter adjustments and equipment maintenance.

[0067] The entire parameter calibration process achieves self-optimization of the inspection system by dynamically linking defect identification results with inspection parameters. It not only promptly detects misjudgments due to parameter drift but also, through the accumulation and analysis of historical data, predicts potential parameter anomalies and enables proactive calibration, ensuring the inspection system is always in optimal working condition. For example, if a certain inspection parameter deviates from the standard value over multiple consecutive time periods, the system automatically issues an early warning and generates an optimal adjustment plan based on historical calibration data, improving calibration efficiency and accuracy. This parameter calibration method, based on actual inspection results, can better adapt to changes in the production environment, such as light source aging and equipment wear, ensuring the accuracy and stability of PCB solder joint defect detection and providing reliable support for production line quality control.

[0068] Example 5: The implementation of the report output module must be based on the mapping index of the adjusted feature fusion set and the defect type. First, the defect type of the solder joint to be inspected is extracted from the mapping index. For example, in the inspection of a certain PCB board, the system determines that the solder joint numbered A12 has a bridging defect, the solder joint numbered B35 has a cold solder joint defect, and the solder joint numbered C7 has an insufficient solder joint defect through the mapping relationship between the feature fusion set and the defect type. The extraction of these defect types is strictly based on the feature matching rules set in the mapping index. For example, a bridging defect corresponds to an abnormal grayscale gradient between solder joints and a connected geometric shape; a cold solder joint defect corresponds to a solder joint area smaller than the threshold and a high grayscale mean; and an insufficient solder joint defect corresponds to a solder joint volume parameter lower than the standard range and a large texture entropy value.

[0069] After extracting defect types, they need to be prioritized by severity. Severity levels are typically determined based on the degree to which the defect impacts the PCB's electrical performance and mechanical strength. For example, fatal defects (such as circuit shorts caused by bridging) have the highest priority, followed by severe defects (such as poor contact caused by cold solder joints), and common defects (such as minor insufficient solder) have the lowest priority. Assuming that in the above example, bridging defects are fatal, cold solder joints are severe, and insufficient solder joints are common, the priority ranking is: A12 solder joint (bridging) > B35 solder joint (cold solder joint) > C7 solder joint (inadequate solder joint).

[0070] After determining the defect handling order for the solder joints to be inspected, the defect type, spatial coordinates, and defect handling order are integrated into a table. The columns in the table typically include fields such as "Solder Joint Number," "Defect Type," "Spatial Coordinates (Row / Column)," "Severity," and "Handling Order." For example, assuming the row coordinates of a PCB are represented by the letters A-Z and the column coordinates by the numbers 1-100, the integrated table is as follows:

[0071] In the table, the spatial coordinates must be consistent with the actual coordinate system of the PCB board. These coordinates are typically determined by the relationship between the inspection region of interest and the pad location established by the image preprocessing module. For example, the spatial coordinates A / 12 for solder joint A12 indicate its location in row A, column 12 of the PCB board. These coordinates are obtained by spatially projecting the row and column coordinates of the inspection region of interest to ensure a one-to-one correspondence with the physical location on the PCB board.

[0072] After integration is complete, the system generates a defect inspection report based on a structured template. The report typically consists of a cover page, table of contents, inspection overview, defect details table, and action suggestions. The cover page includes information such as the report title, inspection date, and inspection object (e.g., PCB model and batch number). The inspection overview briefly describes the inspection purpose, scope, and methods. The defect details table is the integrated form described above and can be expanded to multiple pages as needed. The action suggestions section provides specific remedial measures for different defect types, such as using a soldering iron to remove excess solder for bridging defects, resoldering for cold joint defects, and adding solder for insufficient solder defects.

[0073] For example, a PCB inspection report might include the following: "PCB Solder Joint Defect Inspection Report," "Inspection Date: June 22, 2025," and "Inspection Object: XX Model PCB (Batch Number: PCBA2025062201)." The inspection overview section explains that an image recognition-based inspection system automated inspection of 500 solder joints on the PCB identified three defects. In addition to the three solder joints mentioned above, any other defects must be listed individually in the defect details table. The spatial coordinates of each solder joint must be accurately mapped to the PCB grid to facilitate quick location by production personnel.

[0074] The handling recommendations section details the steps and precautions for each defect type. For example, for the A12 solder joint bridging defect, the recommendation is to "use a constant-temperature soldering iron (temperature setting 350±10°C) with a fine-pointed soldering iron tip to gently stroke along the bridging area between the solder joints to remove excess solder. Avoid pulling hard on the PCB board during this operation." For the B35 solder joint cold solder defect, the recommendation is to "first apply flux to the solder joint surface, then heat the solder joint with the soldering iron until the solder melts, ensuring that the solder fully soaks the pad and pin. After cooling, check the solder joint for gloss and fullness." For the C7 solder joint insufficient solder defect, the recommendation is to "use fine-gauge solder wire (0.5mm diameter) to add solder to the edge of the solder joint, controlling the heating time to 2-3 seconds to avoid excessive soldering and causing new bridging defects."

[0075] During report generation, the system automatically calculates defect distribution characteristics, such as the percentage of different defect types and their regional distribution on the PCB. For example, the three defects in the example above are located in the upper left corner (A12), middle (B35), and lower right corner (C7) of the PCB, respectively, with no apparent concentrated distribution trend. This suggests that the defects may be caused by random factors. If multiple defects are concentrated in a single area, it is necessary to investigate whether there are any abnormalities in the soldering process parameters in that area, such as uneven soldering station temperature or inconsistent solder paste thickness.

[0076] The report format can be customized to meet the manufacturer's specific needs, such as adding a company logo, a field for inspector signatures, and an audit process. For batch-tested PCBs, the report can also include statistical charts, such as a pie chart showing defect types and a trend chart showing defect counts by inspection period, to help managers gain an intuitive understanding of product quality. For example, a pie chart showing defect types for a particular batch of PCBs reveals that bridging defects account for 40%, cold solder joints for 30%, insufficient solder for 20%, and other defects for 10%. This information can help managers identify potential issues in the soldering process, such as excessive solder temperature (causing bridging) or insufficient solder supply (resulting in insufficient solder), and allow them to adjust production parameters accordingly.

[0077] The generated defect detection reports are automatically sent through the system to relevant departments, such as Production, Quality, and Process Engineering. These departments then collaborate to address the defects based on the report content. The Production Department is responsible for repairing defective solder joints according to recommended treatments, while the Quality Department reviews the repair results. The Process Engineering Department analyzes potential production process issues based on defect distribution and type, and proposes process optimization solutions. For example, if a high proportion of bridging defects persists across multiple batches of PCBs, the Process Engineering Department can adjust parameters such as solder wave height and transport speed, and verify the optimization results through subsequent testing.

[0078] The entire report output module implementation process, from defect information extraction to structured report generation, forms a complete quality traceability chain. Through precise defect location, clear severity classification, and specific handling recommendations, it not only provides direct guidance for defect repair at the production site, but also provides data support for process improvement and quality enhancement. For example, after using the reports generated by this system, an electronics factory discovered that the cold solder joint defects of a certain model of PCB board were concentrated on the solder joints of specific components. Analysis revealed that the oxide layer on the component pins had not been cleaned properly. Subsequent pretreatment steps were added to significantly reduce the number of such defects. This data-driven quality control model can effectively improve the production yield of PCB boards, reduce production costs, and enhance the company's competitiveness.

[0079] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0080] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A PCB solder joint defect detection system based on image recognition, characterized in that: include: The image preprocessing module is used to obtain the original image stream of the solder joints to be inspected on the PCB board and divide the inspection area of ​​interest of the solder joints to be inspected according to the process parameters of the PCB board; The feature fusion module is used to extract multimodal features from the original image stream, obtain the geometric features, grayscale distribution features, and texture gradient features of the solder joints to be detected, and form a solder joint feature fusion set; The defect discrimination module is used to obtain the discrimination benchmark of the solder joint feature fusion set, establish the mapping index between the feature fusion set and the defect type, calculate the matching coefficient between the solder joint feature fusion set and the standard sample, and obtain the defect type discrimination result; The parameter calibration module is used to verify the consistency of the inspection parameters of the solder joint to be inspected based on the defect type identification results, identify abnormal offsets in the inspection parameters, and adjust the mapping index between the feature fusion set and the defect type based on the abnormal offsets; The report output module is used to extract the defect type and spatial coordinates of the solder joint to be inspected based on the adjusted feature fusion set and the mapping index of the defect type, and generate a defect inspection report according to the structured template.

2. The PCB solder joint defect detection system based on image recognition according to claim 1 is characterized in that: The implementation of the image preprocessing module includes: For any solder joint to be inspected on the PCB board, an industrial line scan camera matching the process parameters is called; The original image stream of the solder joint is collected by an industrial line scan camera, and the image contrast is enhanced by using histogram equalization operation; The edge gradient value of the solder joint area in the enhanced image is identified, and the continuous pixel group with gradient value greater than the threshold is extracted as the detection area of ​​interest of the solder joint to be detected.

3. The PCB solder joint defect detection system based on image recognition according to claim 2 is characterized in that: Methods for extracting a continuous pixel group whose gradient value is greater than a threshold value also include: Convert the enhanced image into HSL color space and calculate the brightness component of each pixel; Extract pixel points whose brightness components are within a preset range as candidate edge points, and connect adjacent candidate edge points to form a closed area; Taking the closed area as the boundary, the pixel matrix of the corresponding area in the HSL image is intercepted as the detection area of ​​interest of the solder joint to be detected.

4. The PCB solder joint defect detection system based on image recognition according to claim 1, characterized in that: The implementation method of dividing the inspection area of ​​interest of the solder joint to be inspected also includes: Mark the coordinates of the detection area of ​​interest, analyze the row and column coordinates of the area on the PCB board, perform spatial projection of the detection area of ​​interest according to the row and column coordinates, and establish a positioning association relationship between the detection area of ​​interest and the pad position on the PCB board.

5. The PCB solder joint defect detection system based on image recognition according to claim 1, characterized in that: The implementation methods of the feature fusion module include: Invoking a geometric morphology algorithm, a grayscale statistics algorithm, and a texture analysis algorithm for the solder joint to be inspected to generate multiple unfused feature extraction results, where the unfused feature extraction results represent geometric parameters, grayscale parameters, and texture parameters that are not associated with the defect type; It is determined whether multiple unfused feature extraction results are valid feature subsets. If they are valid feature subsets, the valid feature subsets are merged into a solder joint feature fusion set.

6. The PCB solder joint defect detection system based on image recognition according to claim 3, characterized in that: The implementation method of establishing a mapping index between a feature fusion set and a defect type includes: using the geometric feature description, grayscale distribution value and texture gradient pattern in the weld feature fusion set, combined with the definition parameters of the defect type and the identification code of the defect type, to establish a mapping index between the feature fusion set and the defect type.

7. The PCB solder joint defect detection system based on image recognition according to claim 1, characterized in that: The implementation methods of the defect identification module include: Cluster the solder joint feature fusion set and defect types according to defect shape, defect size, and impact level, and set the main cluster center after clustering as the feature discrimination benchmark; Extracting the core feature parameters of the feature discrimination benchmark, calculating the parameter similarity between the core feature parameters, and setting a feature sequence related to the parameter similarity between the core feature parameters; Using the feature sequence related to the parameter similarity between each core feature parameter, the defect features in the sequence are extracted, and the longest common feature subsequence between each defect feature is set, and the matching value of the longest common feature subsequence is set as the defect feature matching coefficient; The defect feature matching coefficients between the defect features are distributed according to the occurrence frequency of each defect feature to set the defect type discrimination result.

8. The PCB solder joint defect detection system based on image recognition according to claim 7, characterized in that: The implementation of the parameter calibration module includes: Extracting the frequency distribution of each defect feature from the defect type discrimination result; setting a calibration path for the defect type discrimination result according to the detection period corresponding to the frequency distribution of each defect feature; Perform curve fitting on the calibration path of each defect feature in the defect type discrimination result to obtain a fitted calibration path, and set the probability value of the fitted path in each detection period as the discrimination confidence of the defect type discrimination result; The confidence level of the defect type identification result is compared with the standard reference value of the detection parameter, the existing offset difference is marked, and the detection parameter is corrected according to the offset difference to complete the calibration of the detection parameter.

9. The PCB solder joint defect detection system based on image recognition according to claim 1, characterized in that: The report output module is implemented as follows: Based on the adjusted feature fusion set and the mapping index of the defect type, the defect type of the solder joint to be inspected is extracted, and the defect type is prioritized according to the severity level of the defect type to obtain the defect processing order of the solder joint to be inspected; the defect type, spatial coordinates and defect processing order are integrated in a tabular form to generate a defect detection report.

10. A PCB solder joint defect detection method based on image recognition, applied to the PCB solder joint defect detection system based on image recognition according to any one of claims 1 to 9, characterized in that: The following steps are involved: Obtain the original image stream of the solder joints to be inspected on the PCB board, and divide the inspection area of ​​interest of the solder joints to be inspected according to the process parameters of the PCB board; Perform multimodal feature extraction on the original image stream to obtain the geometric features, grayscale distribution features, and texture gradient features of the solder joints to be detected, forming a solder joint feature fusion set; Obtain the discrimination benchmark of the solder joint feature fusion set, establish a mapping index between the feature fusion set and the defect type, calculate the matching coefficient between the solder joint feature fusion set and the standard sample, and obtain the defect type discrimination result; Using the defect type identification results as a reference, the consistency of the inspection parameters of the solder joint to be inspected is verified, abnormal offsets in the inspection parameters are identified, and the mapping index between the feature fusion set and the defect type is adjusted based on the abnormal offsets. Based on the adjusted mapping index, the defect type and spatial coordinates of the solder joint to be inspected are extracted, and a defect inspection report is generated according to the structured template.

Images