Circuit board defect detection system and method

Abstract

The invention relates to the technical field of image processing, in particular to a circuit board defect detection system and method. According to the method, a plurality of detection sub-regions are obtained through division, and spectral image parameters, circuit board design information and process execution historical data are obtained based on circuit board feature information of each region; further obtaining pixel intensity ratio anomaly, function influence degree and historical defect frequency factors, and performing coupling analysis to obtain circuit board region evaluation information to realize region importance grading; the resource waste caused by full-region non-difference detection is avoided, the calculation load is greatly reduced, the detection delay is shortened, and the real-time requirement of high-speed production is met; according to the method, lightweight feature extraction is carried out on multispectral imaging information to obtain a fused multispectral image, the defect detection rate of a core area is guaranteed, the process of a non-core area is simplified through lightweight processing and low-standard detection, the dynamic balance of detection precision and efficiency is improved, and the overall detection quality is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a circuit board defect detection system and method. Background Technology

[0002] Circuit boards (PCBs) are the core foundation of the electronics industry's evolution. Their technological advancements not only support the miniaturization, high performance, and low power consumption upgrades of electronic devices, adapting to the demands of emerging fields such as 5G communication, artificial intelligence, new energy vehicles, and aerospace, but also drive functional innovation and scenario expansion of terminal products. From consumer electronics to medical devices and industrial control, the large-scale application of high-quality PCBs directly determines the reliability, stability, and lifespan of electronic devices, serving as a key support for ensuring the security of the industrial chain and supply chain, breaking through core technological bottlenecks, and enhancing the global competitiveness of my country's electronics manufacturing industry. PCB defect detection is a core link in electronic manufacturing quality control, and its technological upgrades are of strategic significance for improving industrial quality and efficiency. Currently, full image processing for PCB defect detection consumes significant computational resources, wastes computation in irrelevant areas, and results in high detection latency, making it difficult to meet the real-time requirements of high-speed production. Summary of the Invention

[0003] This invention provides a circuit board defect detection system and method to solve the above-mentioned technical problems.

[0004] The first aspect of the present invention provides a circuit board defect detection system, including a spectral image input module, a spectral feature extraction module, a regional importance calculation module, and a precise detection output module.

[0005] The spectral image input module is used to acquire multispectral images of the circuit board, including visible light imaging units and non-visible light imaging units.

[0006] The spectral feature extraction module is used to connect with the spectral image input module to acquire multispectral images and extract circuit board feature information based on the multispectral images.

[0007] The regional importance calculation module is used to obtain circuit board feature information and perform regional importance calculation analysis on the circuit board feature information to obtain circuit board regional evaluation information.

[0008] As a further improvement of the present invention, the regional importance calculation and analysis of the circuit board feature information is performed, and the specific analysis method is as follows: The area to be inspected on the circuit board is divided into multiple inspection sub-regions according to a preset regional interval. Spectral image parameters, circuit board design information, and process execution history data are obtained based on the circuit board feature information of each inspection sub-region. Spectral image parameters are analyzed to obtain pixel intensity ratio anomaly. Circuit board design information is analyzed to obtain functional impact. Process execution history data is analyzed to obtain historical defect frequency factor. Pixel intensity ratio anomaly, functional impact, and historical defect frequency factor are coupled and analyzed to obtain circuit board area evaluation information. Furthermore, the pixel intensity ratio anomaly is obtained by analyzing the spectral image parameters. The specific analysis content is as follows: Two spectral channels are randomly selected based on the spectral image parameters, and their corresponding gray values ​​are obtained. Then, the joint probability distribution of the gray values ​​of the two spectral channels is estimated to obtain the joint probability distribution P(A, B), where A and B both belong to the spectral channels. The spectral channels include the imaging channels corresponding to the visible light imaging unit and the non-visible light imaging unit, that is, at least the visible light, infrared and thermal spectral channels, denoted as VIS, NIR and TP, respectively. Based on the database, normal sample data of a preset number of circuit boards of the same model are obtained, and the average mutual information value of each spectral channel pair is obtained, denoted as... k belongs to the spectral channel pair; the measured mutual information of the current detection sub-region is obtained based on the spectral image parameters, and the difference between it and the average mutual information value of the corresponding samples is calculated to obtain the mutual information deviation value. The mutual information deviation value corresponding to each normal sample data is calculated separately, that is, by the formula Calculate and output the mutual information deviation value Then, based on normal sample data, the maximum and minimum mutual information deviation values ​​are obtained. The three-channel pixel intensity ratio vector is obtained by calculating the gray values ​​of each spectral channel in the detection sub-region using the measured pixel gray values, i.e., by using the formula... Calculate and output the three-channel pixel intensity ratio vector ,in The corresponding gray values ​​of each spectral channel are obtained; the statistical distance between the three-channel pixel intensity ratio vector and the three-channel pixel intensity ratio vector of normal sample data is obtained based on the Mahalanobis distance method, and recorded as the regional spectral ratio vector; historical abnormal samples are obtained based on the database, and the maximum Mahalanobis distance of the historical abnormal samples is obtained; the pixel intensity ratio anomaly degree is calculated and output using the preset mutual information deviation and Mahalanobis distance coupling formula.

[0009] Furthermore, the functional impact is obtained by analyzing the circuit board design information. The specific analysis content is as follows: Based on the circuit board design information, the electrical nodes of the circuit board are obtained, and each electrical node is mapped to a node in a preset electrical topology network, i.e., an electrical topology node, denoted as . ; Betweenness centrality measures the frequency at which each electrical topology node acts as a shortest path hub in the topology network and outputs it as the centrality frequency. The specific calculation principle is as follows: calculate the shortest path length for all node pairs (s, t) in the topology network, where s and t are both electrical topology nodes; substitute this length into the preset centrality frequency calculation formula. Calculate the output central frequency ;in Let be the total number of shortest paths from s to t. For the nodes Find the shortest path; and normalize the centrality frequency so that its value range is [0, 1]; Based on the circuit board design information, obtain the set of core functional nodes of the circuit board; obtain the electrical topology nodes. The shortest path length to each core functional node f; Obtain the core node scores corresponding to the preset severity, occurrence, and detectability of the core nodes; and assign a corresponding core node scoring mechanism, i.e., the core node score takes the value of [1, 10], and the higher the score, the higher the probability; calculate the risk priority through the preset risk priority calculation formula, and input it into the preset fault impact weight calculation formula to calculate and output the fault impact weight; The central frequency, the shortest path length from the electrical topology node to the core functional node, and the fault impact weight are input into the preset functional impact calculation formula to calculate and output the functional impact.

[0010] Furthermore, the historical defect frequency factor is obtained by analyzing the historical process execution data. The specific analysis content is as follows: Based on the historical process execution data, the defect identifiers for each historical inspection in each inspection sub-region are obtained and denoted as follows: This refers to the defect identifier corresponding to the i-th historical inspection, which takes a value of 0 or 1. A value of 0 indicates no defect, and a value of 1 indicates a defect. The historical inspection time corresponding to each historical inspection is also obtained. and current testing time ; The process improvement half-life Th ​​of the circuit board is obtained based on historical process execution data; and the process correlation coefficients corresponding to each historical test are obtained. ; Using a pre-set historical defect correction calculation formula Calculate and output the historical defect frequency factor Where sn is the total number of historical test batches of the same type of circuit board.

[0011] Furthermore, the circuit board area evaluation information is obtained by coupling analysis of pixel intensity ratio anomaly, functional impact, and historical defect frequency factor. The specific analysis content is as follows: The core parameter vector Hx is obtained by composing a column vector from the pixel intensity ratio anomaly, functional impact, and historical defect frequency factor. Based on the database, the variance of the parameters of each core parameter vector corresponding to the normal sample is obtained, and then the adaptive weight matrix Wx is obtained through the parameter adaptive weight output algorithm. The parameters of historical defect samples for each core parameter vector are obtained from the database, and the parameters are calculated using a pre-defined Pearson correlation coefficient method. Calculate the correlation matrix of the output parameters. ;in, For historical defect samples and covariance, .

[0012] Based on the preset regional coupling evaluation formula The system calculates and outputs a region importance index (PS). When the region importance index exceeds a preset importance threshold, the corresponding detection sub-region is marked as a precise detection region. When the region importance index is less than a preset lower importance threshold, the corresponding detection sub-region is marked as a low detection demand region. Then, the region evaluation information of the detection sub-regions falling between the lower and higher importance thresholds is marked as normal detection regions. The region evaluation information of all detection sub-regions is then aggregated to obtain the circuit board region evaluation information. For Gaussian radial basis functions, This is the width parameter for the RBF.

[0013] The precision detection output module obtains defect detection information by performing precision defect detection based on the circuit board area evaluation information. Specifically, it performs high-precision defect detection on areas where the circuit board area evaluation information is for precision detection, standardized defect detection on areas for routine detection, and low-standard defect detection on areas with low detection requirements. High-precision defect detection, standardized defect detection, and low-standard defect detection are all preset defect detection process standards.

[0014] A second aspect of the present invention provides a method for detecting defects in a circuit board, comprising: Step 1: Multispectral imaging information is acquired by the visible light imaging unit and non-visible light imaging unit of the spectral image input module.

[0015] Step 2: Perform lightweight feature extraction on the multispectral imaging information to obtain a fused multispectral image.

[0016] As a further improvement to the present invention, lightweight feature extraction is performed on multispectral imaging information, as follows: Based on the circuit board design information, the functional units of the circuit board are obtained. The multispectral image is then segmented according to the functional units to obtain functional segmentation regions, and these regions are then aggregated to obtain a set of functional segmentation regions. , This represents the total number of functionally segmented regions. Obtain the preset imaging quality score cx and defect sensitivity mg corresponding to each functional segmentation region, and assign a corresponding segmentation region score, i.e., the segmentation region score is [1, 10], with higher scores indicating higher probabilities; perform spectral channel quality scoring TQ on each functional segmentation region, and denote the spectral channel as GP. The spectral channels include visible light, infrared, and thermal channels, respectively denoted as... The quality scoring formula is: ; Optimal spectral channel selection is performed based on spectral channel quality scores, i.e., using a pre-defined channel selection formula. Functional segmentation areas are obtained by selecting channels. Optimal spectral channel Based on the optimal spectral channels of each functional segmentation region, the corresponding regional channel imaging information is obtained and marked as the optimal regional imaging information. The optimal imaging information of each functional segmentation region is then combined to obtain a fused multispectral image.

[0017] Step 3: Perform precise defect detection analysis by combining the multispectral image input region importance calculation module and the precise detection output module.

[0018] The beneficial effects of the technical solution provided by this invention compared with the prior art are as follows: 1. This invention divides the area to be inspected on a circuit board into multiple inspection sub-regions according to a preset regional interval; based on the circuit board feature information of each inspection sub-region, spectral image parameters, circuit board design information, and process execution history data are obtained; then, pixel intensity ratio anomaly, functional impact, and historical defect frequency factor are obtained, and coupled analysis is performed to obtain circuit board area evaluation information, realizing the classification of regional importance; avoiding the waste of resources caused by indiscriminate inspection of the entire area, significantly reducing the computational load, shortening the inspection delay, and meeting the real-time requirements of high-speed production.

[0019] 2. This invention acquires multispectral imaging information through the visible light imaging unit and non-visible light imaging unit of the spectral image input module; it then performs lightweight feature extraction on the multispectral imaging information to obtain a fused multispectral image; and finally, it performs precise defect detection analysis by the fused multispectral image input region importance calculation module and the precise detection output module. That is, it ensures the defect detection rate of the core area through optimal spectral channel selection and high-precision detection process, and simplifies the process of non-core areas through lightweight processing and low-standard detection, thereby improving the dynamic balance between detection accuracy and efficiency and improving the overall detection quality. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not deliberately drawn to scale according to the actual size, but are intended to show the main idea of ​​this application.

[0021] Figure 1 This is a schematic diagram of the circuit board defect detection system of the present invention; Figure 2 This is a flowchart of the circuit board defect detection method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 In one embodiment of the circuit board defect detection system of the present invention, the following is included: The spectral image input module acquires multispectral images of the circuit board, including a visible light imaging unit and a non-visible light imaging unit, to obtain image information of the circuit board in different spectral frequency bands and obtain multispectral images.

[0024] The spectral feature extraction module is connected to the spectral image input module to acquire multispectral images, and the circuit board feature information is extracted based on the multispectral images.

[0025] The regional importance calculation module acquires circuit board feature information and performs regional importance calculation analysis on the circuit board feature information to obtain circuit board regional evaluation information.

[0026] The specific analysis method for calculating the regional importance of circuit board feature information is as follows: The area to be tested on the circuit board is divided into multiple sub-regions according to a preset regional interval; spectral image parameters, circuit board design information, and process execution history data are obtained based on the circuit board feature information of each sub-region. The pixel intensity ratio anomaly is obtained by analyzing the spectral image parameters. Specifically, two spectral channels are randomly selected based on the spectral image parameters, and their corresponding gray values ​​are obtained. Then, the joint probability distribution of the gray values ​​of the two spectral channels is estimated to obtain the joint probability distribution P(A, B), where A and B both belong to the spectral channels. The spectral channels include the imaging channels corresponding to the visible light imaging unit and the non-visible light imaging unit, that is, at least the visible light, infrared, and thermal spectral channels, denoted as VIS, NIR, and TP, respectively. The joint probability distribution estimation of the gray values ​​of the two spectral channels is carried out by using the conventional kernel density estimation method in this field. Based on the gray value sample set of the two channels in the evaluation area, a two-dimensional joint probability density function is constructed. The edge probability distribution is calculated by integrating the joint probability distribution over the dimension of a single channel, or by directly estimating the gray value of a single channel using the histogram statistical method.

[0027] Based on the database, normal sample data of a preset number of circuit boards of the same model are obtained, and the average mutual information value of each spectral channel pair (i.e., VIS-NIR, VIS-TP, and NIR-TP) is obtained, denoted as... k belongs to the spectral channel pair; the measured mutual information of the current detection sub-region is obtained based on the spectral image parameters, and the difference between it and the average mutual information value of the corresponding samples is calculated to obtain the mutual information deviation value. The mutual information deviation value corresponding to each normal sample data is calculated separately, that is, by the formula Calculate and output the mutual information deviation value Then, based on normal sample data, the maximum mutual information deviation value and the minimum mutual information deviation value are obtained; the normal sample data is collected, for example, by selecting 1,000 circuit boards of the same model and batch that have been confirmed to be defect-free by factory inspection, and acquiring grayscale images of each imaging channel through the same multispectral imaging device as the detection system of this invention.

[0028] The three-channel pixel intensity ratio vector is obtained by calculating the gray values ​​of each spectral channel in the detection sub-region using the measured pixel gray values, i.e., by using the formula... Calculate and output the three-channel pixel intensity ratio vector ,in These correspond to the grayscale values ​​of each spectral channel; the statistical distance between the three-channel pixel intensity ratio vector and the three-channel pixel intensity ratio vector of the normal sample data is obtained based on the Mahalanobis distance method, and is denoted as the region spectral ratio vector (its calculation principle is based on the Mahalanobis distance formula). Calculate the spectral ratio vector of the output region. ;in, These are the mean vector and covariance matrix of the intensity ratio vectors of the three channels of normal sample data, respectively. Historical abnormal samples are obtained from the database, and the maximum Mahalanobis distance of the historical abnormal samples is obtained. The historical abnormal samples are defect cases of the same type of circuit board recorded by the factory system, including but not limited to typical defects such as cold solder joints, delamination and cracks, to ensure coverage of major abnormal scenarios.

[0029] Using a pre-defined coupling formula between mutual information bias and Mahalanobis distance Calculate the anomaly of the output pixel intensity ratio. ;in, These represent the maximum mutual information deviation value and the minimum mutual information deviation value, respectively. This represents the maximum Mahalanobis distance.

[0030] The functional impact is obtained by analyzing the circuit board design information. Specifically, the electrical nodes of the circuit board are obtained based on the circuit board design information, and each electrical node is mapped to a node in a preset electrical topology network (where the electrical topology network is defined as nodes = functional areas of the circuit board, edges = wiring connections between detection sub-regions, and edge weights = signal transmission impedance of the wiring). This node is denoted as _____. For example, the power module pads correspond to The high-speed data bus wiring corresponds to The mapping is based on the circuit netlist data extracted from the circuit board design file. The electrical nodes such as component pins, pads, and wiring segments recorded in the netlist correspond one-to-one with the nodes of the topology network. The mapping rules are implemented by the conventional netlist parsing algorithm in this field. The frequency at which each electrical topology node acts as a shortest path hub in the topology network is measured by betweenness centrality and output as the centrality frequency. The calculation principle is as follows: calculate the shortest path length for all node pairs (s, t) in the topology network, where s and t are both electrical topology nodes; substitute this length into the preset centrality frequency calculation formula. Calculate the output central frequency ;in Let be the total number of shortest paths from s to t. For the nodes Find the shortest path; and normalize the centrality frequency so that its value range is [0, 1].

[0031] Based on the circuit board design information, the set of core functional nodes of the circuit board is obtained. For example, the failure of core nodes such as the CPU core power supply terminal, clock crystal oscillator terminal, and high-speed data interface terminal will directly lead to the loss of overall system functionality; electrical topology nodes are obtained. The shortest path length to each core functional node f; The system acquires core node scores based on preset severity (the extent to which a defect at the core node affects the entire circuit board), occurrence (the probability of a defect occurring at the core node during normal production), and detectability (the probability that existing detection methods can detect the defect after it occurs). Each score is assigned a corresponding core node scoring mechanism, with scores ranging from [1, 10], where higher scores indicate higher probability. The risk priority RPN is calculated using a preset risk priority calculation formula: RPN = Severity × Occurrence × Detectability. Finally, the system uses a fault impact weighting calculation formula... Calculate and output the fault impact weights. ;in, Risk priority for core functional nodes This is the highest risk priority for all core nodes.

[0032] The central frequency, the shortest path length from the electrical topology node to the core functional node, and the fault impact weight are input into the preset functional impact calculation formula. Perform calculations to output the influence of the function. ;in, Electrical topology node The shortest path length to the core functional node f; TJ represents the electrical topology node.

[0033] The historical defect frequency factor is obtained by analyzing the historical process execution data. Specifically, it involves obtaining the defect identifiers for each historical inspection in each inspection sub-region based on the historical process execution data, denoted as... This refers to the defect identifier corresponding to the i-th historical inspection, which takes a value of 0 or 1. A value of 0 indicates no defect, and a value of 1 indicates a defect. The historical inspection time corresponding to each historical inspection is also obtained. and current testing time .

[0034] The process improvement half-life Th ​​of the circuit board is obtained based on historical process execution data (an empirical parameter set based on process improvement verification data, representing the time required for historical defect risk to decrease to 50% after process improvement. For example, the improvement half-life of solder paste printing process is 15 days, meaning that the defect risk weight 15 days ago decreases to half of the current value); and the process correlation coefficient corresponding to each historical inspection is obtained. Its value is set based on the process parameters of the circuit board consumables. For the same batch of process parameters, a value of 1.2 is used; for different batches, a value of 1.0 is used. For example, when the process parameters of the solder paste type, reflow temperature profile, and stencil specifications of the circuit board are consistent across batches, a value of 1.0 is used. =1.2, and vice versa =1.0.

[0035] Using a pre-set historical defect correction calculation formula Calculate and output the historical defect frequency factor Where sn is the total number of historical test batches of the same type of circuit board.

[0036] The circuit board area evaluation information is obtained by coupling analysis of pixel intensity ratio anomaly, functional impact, and historical defect frequency factor. Specifically: The core parameter vector Hx is obtained by assembling a column vector from the pixel intensity ratio anomaly, functional impact, and historical defect frequency factor. ; Based on the database, the variance of the parameters corresponding to the normal samples for each core parameter vector is obtained, and then the adaptive weight matrix Wx is obtained through a parameter adaptive weight output algorithm; that is... The variance of the parameters for the normal sample is That is, the weight value is... ; The parameters of historical defect samples for each core parameter vector are obtained from the database, and the parameters are calculated using a pre-defined Pearson correlation coefficient method. Calculate the correlation matrix of the output parameters. ;in, For historical defect samples and covariance, .

[0037] Based on the preset regional coupling evaluation formula The system calculates and outputs a region importance index (PS). When the region importance index exceeds a preset importance threshold, the corresponding detection sub-region is marked as a precise detection region. When the region importance index is less than a preset lower importance threshold, the corresponding detection sub-region is marked as a low detection demand region. Then, the region evaluation information of the detection sub-regions falling between the lower and higher importance thresholds is marked as normal detection regions. The region evaluation information of all detection sub-regions is then aggregated to obtain the circuit board region evaluation information. For Gaussian radial basis functions, The RBF width parameter is used to select normal samples. The standard deviation is used to ensure that the PS in the normal region approaches 0 and the PS in the abnormal region approaches 1; the Gaussian radial basis function is used as a nonlinear activation function to achieve the effect of exponential compression in the normal region and exponential amplification in the abnormal region, that is, the coupling intermediate value in the normal region. Approaching 0 It is close to 1, but combined with parameter coupling logic, the normal region The actual value approaches 0, and the final PS value approaches 0; the coupling intermediate value in the abnormal region. Stay away from 0, The value drops rapidly, but through the symbol design of parameter coupling, the PS of the abnormal region eventually approaches 1; PS∈[0,1], the larger the value, the higher the importance of the region and the more refined the detection is required.

[0038] The precision detection output module performs precision defect detection based on the circuit board area assessment information to obtain defect detection information. Specifically, it performs high-precision defect detection on areas where the circuit board area assessment information is a precision detection area, standardized defect detection on areas with normal detection requirements, and low-standard defect detection on areas with low detection requirements. High-precision defect detection, standardized defect detection, and low-standard defect detection are all preset defect detection process standards.

[0039] To facilitate calculations, all index data involved in the calculations in this embodiment of the invention have undergone data preprocessing to eliminate the influence of dimensions. The specific methods for eliminating the influence of dimensions are well-known to those skilled in the art and are not limited here. Please see Figure 2 The present invention also provides a method for detecting circuit board defects, including: Step 1, Multispectral Image Acquisition: Multispectral imaging information is acquired through the visible light imaging unit and non-visible light imaging unit of the spectral image input module.

[0040] Step 2, Multispectral Feature Extraction and Analysis: Lightweight feature extraction is performed on the multispectral imaging information to obtain a fused multispectral image.

[0041] Lightweight feature extraction of multispectral imaging information is performed as follows: Based on the circuit board design information, the functional units of the circuit board are obtained. The multispectral image is then segmented according to the functional units to obtain functional segmentation regions, and these regions are then aggregated to obtain a set of functional segmentation regions. , This represents the total number of functionally segmented regions. Obtain the preset imaging quality score cx and defect sensitivity mg corresponding to each functional segmentation region, and assign a corresponding segmentation region score, i.e., the segmentation region score is [1, 10], with higher scores indicating higher probabilities; perform spectral channel quality scoring TQ on each functional segmentation region, and denote the spectral channel as GP. The spectral channels include visible light, infrared, and thermal channels, respectively denoted as... The quality scoring formula is: ;right The quality score is based on the contrast of gray values ​​in the spectral channel imaging to segment and score the regions. The quality score is based on the standard deviation of reflectance of the infrared spectral channels to segment and score different regions. The quality score is based on the temperature uniformity of the thermal spectrum channel, and the score is divided into regions.

[0042] Optimal spectral channel selection is performed based on spectral channel quality scores, i.e., using a pre-defined channel selection formula. Functional segmentation areas are obtained by selecting channels. Optimal spectral channel Based on the optimal spectral channels of each functional segmentation region, the corresponding regional channel imaging information is obtained and marked as the optimal regional imaging information. The optimal imaging information of each functional segmentation region is then combined to obtain a fused multispectral image.

[0043] Step 3, Refined Detection: The importance calculation module for the input region of the fused multispectral image and the precise detection output module are used to perform precise defect detection and analysis.

[0044] To facilitate calculations, all index data involved in the calculations in this embodiment of the invention have undergone data preprocessing to eliminate the influence of dimensions. The specific methods for eliminating the influence of dimensions are well-known to those skilled in the art and are not limited here.

[0045] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A circuit board defect detection system, characterized in that, include: The spectral image input module is used to acquire multispectral images of the circuit board, including a visible light imaging unit and a non-visible light imaging unit; The spectral feature extraction module is used to connect with the spectral image input module to acquire multispectral images and extract circuit board feature information based on the multispectral images; The regional importance calculation module is used to acquire circuit board feature information and perform regional importance calculation analysis on the circuit board feature information to obtain circuit board regional evaluation information. The specific analysis method for calculating the regional importance of circuit board feature information is as follows: The area to be tested on the circuit board is divided into multiple sub-regions according to a preset regional interval; spectral image parameters, circuit board design information, and process execution history data are obtained based on the circuit board feature information of each sub-region; the pixel intensity ratio anomaly is obtained by analyzing the spectral image parameters; and the functional impact is obtained by analyzing the circuit board design information. Historical defect frequency factor is obtained by analyzing historical process execution data; The circuit board area evaluation information is obtained by coupling analysis of pixel intensity ratio anomaly, functional impact and historical defect frequency factor. The precision detection output module performs precise defect detection based on the circuit board area evaluation information to obtain defect detection information.

2. The circuit board defect detection system according to claim 1, characterized in that, The pixel intensity ratio anomaly is obtained by analyzing the spectral image parameters. The specific analysis content is as follows: Two spectral channels are randomly selected based on the spectral image parameters, and their corresponding gray values ​​are obtained. Then, the joint probability distribution of the gray values ​​of the two spectral channels is estimated to obtain the joint probability distribution P(A, B), where A and B both belong to the spectral channels. The spectral channels include the imaging channels corresponding to the visible light imaging unit and the non-visible light imaging unit, that is, at least the visible light, infrared and thermal spectral channels, denoted as VIS, NIR and TP, respectively. The database is used to obtain a preset number of normal sample data for the same model of circuit board, and the average mutual information value of each spectral channel pair is obtained. The measured mutual information of the current detection sub-region is obtained based on the spectral image parameters, and the difference between the measured mutual information and the average mutual information value of the corresponding sample is calculated to obtain the mutual information deviation value. The mutual information deviation value corresponding to each normal sample data is calculated respectively. Then, the maximum mutual information deviation value and the minimum mutual information deviation value are obtained based on the normal sample data. The gray values ​​of each spectral channel in the detection sub-region are calculated from the measured pixel gray values ​​to obtain the three-channel pixel intensity ratio vector; the statistical distance between the three-channel pixel intensity ratio vector and the three-channel pixel intensity ratio vector of the normal sample data is obtained based on the Mahalanobis distance method, and recorded as the region spectral ratio vector; Historical anomaly samples are obtained from the database, and the maximum Mahalanobis distance of the historical anomaly samples is obtained; the pixel intensity ratio anomaly degree is calculated using a preset formula that couples mutual information deviation with Mahalanobis distance.

3. The circuit board defect detection system according to claim 2, characterized in that, The functional impact is obtained by analyzing the circuit board design information, and the specific analysis content is as follows: Based on the circuit board design information, the electrical nodes of the circuit board are obtained, and each electrical node is mapped to a node of a preset electrical topology network, i.e., an electrical topology node, denoted as . The frequency at which each electrical topology node acts as a shortest path hub in the topology network is measured using betweenness centrality, and the output frequency is the centrality frequency; the set of core functional nodes of the circuit board is obtained based on the circuit board design information; and the electrical topology nodes are acquired. Find the shortest path length to each core functional node f; obtain the core node scores corresponding to the preset severity, occurrence, and detectability of the core nodes; And assign a corresponding core node scoring mechanism; The risk priority is calculated using a preset risk priority calculation formula, and then input into a preset fault impact weight calculation formula to calculate and output the fault impact weight. The central frequency, the shortest path length from the electrical topology node to the core functional node, and the fault impact weight are input into the preset functional impact calculation formula to calculate and output the functional impact.

4. The circuit board defect detection system according to claim 3, characterized in that, The historical defect frequency factor is obtained by analyzing the historical data of the process execution. The specific analysis content is as follows: Based on the historical process execution data, the defect identifiers for each historical inspection in each inspection sub-region are obtained and denoted as follows: , that is, the defect identifier corresponding to the i-th historical detection, which takes the value of 0 or 1. When the value is 0, there is no defect, and when the value is 1, there is a defect. And obtain the historical detection time corresponding to each historical detection. and current testing time ; The process improvement half-life Th ​​of the circuit board is obtained based on historical process execution data; and the process correlation coefficients corresponding to each historical test are obtained. ; Using a pre-set historical defect correction calculation formula Calculate and output the historical defect frequency factor Where sn is the total number of historical test batches of the same type of circuit board.

5. The circuit board defect detection system according to claim 1, characterized in that, The circuit board area evaluation information is obtained by coupling the pixel intensity ratio anomaly, functional impact, and historical defect frequency factor. The specific analysis content is as follows: The core parameter vector Hx is obtained by composing a column vector from the pixel intensity ratio anomaly, functional impact, and historical defect frequency factor. Based on the database, the variance of the parameters of each core parameter vector corresponding to the normal sample is obtained, and then the adaptive weight matrix Wx is obtained through the parameter adaptive weight output algorithm. The parameters of historical defect samples for each core parameter vector are obtained from the database, and the parameters are calculated using a pre-defined Pearson correlation coefficient method. Calculate the correlation matrix of the output parameters. ;in, For historical defect samples and covariance, ; Based on the preset regional coupling evaluation formula The system calculates and outputs a region importance index (PS). When the region importance index exceeds a preset importance threshold, the corresponding detection sub-region is marked as a precise detection region. When the region importance index is less than a preset lower importance threshold, the corresponding detection sub-region is marked as a low detection demand region. Then, the region evaluation information of the detection sub-regions falling between the lower and higher importance thresholds is marked as normal detection regions. The region evaluation information of all detection sub-regions is then aggregated to obtain the circuit board region evaluation information. For Gaussian radial basis functions, This is the width parameter for the RBF.

6. The circuit board defect detection system according to claim 3, characterized in that, The betweenness centrality measure measures the frequency at which each electrical topology node acts as a shortest path hub in the topology network and outputs the centrality frequency. The specific calculation principle is as follows: calculate the shortest path length for all node pairs (s, t) in the topology network, where s and t are both electrical topology nodes; substitute this length into a preset centrality frequency calculation formula. Perform calculations and output the central frequency. ;in Let be the total number of shortest paths from s to t. For the nodes Find the shortest path; and normalize the centrality frequency so that its value range is [0, 1].

7. The circuit board defect detection system according to claim 1, characterized in that, Based on the circuit board area assessment information, precise defect detection is performed to obtain defect detection information. Specifically, high-precision defect detection is performed on areas where the circuit board area assessment information indicates precise detection, standardized defect detection is performed on areas where normal detection is required, and low-standard defect detection is performed on areas where low detection requirements are required. High-precision defect detection, standardized defect detection, and low-standard defect detection are all preset defect detection process standards.

8. A circuit board defect detection method, applied to the circuit board defect detection system as described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Acquire multispectral imaging information through the visible light imaging unit and non-visible light imaging unit of the spectral image input module; Step 2: Perform lightweight feature extraction on multispectral imaging information and fuse multispectral images; Step 3: Perform precise defect detection analysis by combining the multispectral image input region importance calculation module and the precise detection output module.

9. The circuit board defect detection method according to claim 8, characterized in that, The lightweight feature extraction of multispectral imaging information is specifically performed as follows: Based on the circuit board design information, the functional units of the circuit board are obtained. The multispectral image is then segmented according to the functional units to obtain functional segmentation regions, and these regions are then aggregated to obtain a set of functional segmentation regions. , This represents the total number of functionally segmented regions. Obtain the preset imaging quality score cx and defect sensitivity mg corresponding to each functional segmentation region, and assign a corresponding segmentation region score, i.e., the segmentation region score is [1, 10]. Perform spectral channel quality scoring TQ on each functional segmentation region, and denote the spectral channel as GP. The spectral channels include visible light, infrared, and thermal channels, which are respectively denoted as... ; The quality scoring formula is: ; Optimal spectral channel selection is performed based on spectral channel quality scores, i.e., using a pre-defined channel selection formula. Channel selection is performed to obtain functional segmentation areas. Optimal spectral channel Based on the optimal spectral channels of each functional segmentation region, the corresponding regional channel imaging information is obtained and marked as the optimal regional imaging information. The optimal imaging information of each functional segmentation region is then combined to obtain a fused multispectral image.

Images