Image recognition-based automatic detection method for insufficient solder of solder joints of graphics card

Abstract

The invention relates to the technical field of material detection, and discloses a graphics card welding spot pseudo soldering automatic detection method based on image recognition, and the method comprises the steps: firstly comparing a graphics card visible light image after gray level normalization with an infrared thermal distribution diagram after noise suppression, and obtaining the surface texture and thermodynamic characteristics of a welding spot; and then splicing the two to obtain an associated feature weight matrix, extracting texture and thermodynamic feature vectors of a welding spot region from the associated feature weight matrix, and splicing the texture and thermodynamic feature vectors into a fusion feature matrix according to channel dimensions. And finally, positioning pseudo soldering spot coordinates in combination with the standard soldering spot feature library, and generating a graphics card soldering spot pseudo soldering defect report according to the coordinates and the texture fracture index. According to the process, through multi-feature fusion and comparison, pseudo soldering accurate detection and report generation are achieved; according to the method, the accuracy and the efficiency of insufficient solder detection of the graphics card welding spots based on image recognition can be improved, and a high-precision automatic solution is provided for quality detection of the graphics card welding spots.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material detection, and particularly relates to a graphics card solder joint false welding automatic detection method based on image recognition. BACKGROUND

[0002] The prior art relies on a single type of image information for analysis, and only visible light images are used to extract surface texture features, or only infrared thermal images are used to obtain temperature distribution features. This single feature analysis method cannot fully reflect the complex state of false welding joints. False welding may cause surface texture abnormalities such as cracking and deformation, and may also cause local temperature abnormalities due to poor contact. Relying on only one of these features may result in missed detection or misjudgment, and the subtle differences between normal and false welding joints cannot be accurately distinguished, which limits the detection accuracy.

[0003] The prior art has deficiencies in feature fusion and processing. Although some methods attempt to combine multiple features, they lack an effective dynamic weight distribution mechanism, making it difficult to adjust the weight according to the actual effectiveness of the features, resulting in a large amount of noise interference in the fused feature matrix and weakening of key features. In the feature extraction process, the positioning of the welding point area is not accurate enough, and the deformation state of the welding point cannot be adaptively matched, resulting in a large deviation between the extracted feature vector and the actual welding point state, which ultimately affects the accuracy of false welding coordinate positioning and the reliability of defect reports, making it difficult to meet the needs of high-precision graphics card production detection. SUMMARY

[0004] The present application provides a graphics card solder joint false welding automatic detection method based on image recognition to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides a graphics card solder joint false welding automatic detection method based on image recognition, comprising:

[0006] S1, comparing the visible light image of the graphics card after gray scale normalization and the infrared thermal distribution image of the graphics card after noise suppression position by position to obtain the surface texture features and thermodynamic features of the welding points in the graphics card;

[0007] S2, splicing the surface texture features and the thermodynamic features to obtain the associated feature weight matrix of the graphics card welding points;

[0008] S3, extracting the texture feature vector and the thermodynamic feature vector of the welding point area in the associated feature weight matrix according to the welding point area of the graphics card;

[0009] S4, splicing the texture feature vector and the thermodynamic feature vector in the channel dimension to obtain the fusion feature matrix of the graphics card welding points;

[0010] S5. Locating the coordinates of the cold solder joints in the solder joint area based on the fused feature matrix and the standard solder joint feature library;

[0011] S6. Generate a solder joint defect report of the graphics card according to the coordinates and the texture fracture index of the solder joint.

[0012] In a preferred embodiment, the position-by-position comparison of the grayscale normalized visible light image of the graphics card and the noise-suppressed infrared thermal distribution map of the graphics card to obtain the surface texture characteristics and thermodynamic characteristics of the solder joints in the graphics card includes:

[0013] Extracting the texture key points in the visible light image and the thermal gradient vector field of each pixel point in the infrared thermal distribution map;

[0014] filtering abnormal curvature features in the thermal gradient vector field to obtain an effective solder joint area of ​​the graphics card;

[0015] The texture key points are aligned with the thermal gradient vector field according to the effective solder joint area to obtain surface texture features and thermodynamic features of the solder joints in the graphics card.

[0016] In a preferred embodiment, filtering the abnormal curvature features in the thermal gradient vector field to obtain the effective solder joint area of ​​the graphics card includes:

[0017] Taking the divergence value point exceeding the preset curvature threshold in the thermal gradient vector field as the extreme curvature point of the isothermal line of the graphics card solder joint;

[0018] The area where the extreme point of curvature of the isothermal line overlaps with the key point of the texture is used as the effective soldering point area of ​​the graphics card.

[0019] In a preferred embodiment, the combining of the surface texture features and the thermodynamic features to obtain the associated feature weight matrix of the graphics card solder joints includes:

[0020] splicing the surface texture features and the thermodynamic features along the feature dimension to obtain a high-dimensional tensor of the graphics card solder joint;

[0021] Suppressing high-frequency noise in the high-dimensional tensor;

[0022] Obtaining mask coefficients of the surface texture feature and the thermodynamic feature according to the dynamic distribution of the suppressed high-frequency noise and the weights of the surface texture feature and the thermodynamic feature;

[0023] The mask coefficients are stacked with the spatial feature vectors of the high-dimensional tensor according to pixel positions to obtain a correlation feature weight matrix of the graphics card solder joints.

[0024] In a preferred embodiment, the extracting the texture feature vector and the thermodynamic feature vector of the solder region in the correlation feature weight matrix according to the solder region of the graphics card comprises:

[0025] Driving the sampling points of the correlation feature weight matrix to adaptively offset in the solder deformation direction to obtain a sampling feature map of the graphics card solder;

[0026] Taking the affine transformed sampling coordinate point as a reference, extracting the feature value within the contour boundary of the graphics card solder from the sampling feature map, and taking the feature value as the texture feature vector and the thermodynamic feature vector set of the graphics card solder;

[0027] Selecting the feature vector with a thermodynamic confidence weight greater than a threshold value in the thermodynamic feature vector set as the thermodynamic feature vector of the graphics card solder.

[0028] In a preferred embodiment, the calculation formula of the thermodynamic confidence weight is:

[0029]

[0030] In the formula, is the thermodynamic confidence weight, is the thermal conductivity variation, is the temperature gradient module, is the specific heat capacity, is the material density.

[0031] In a preferred embodiment, the splicing the texture feature vector and the thermodynamic feature vector according to the channel dimension to obtain the fusion feature matrix of the graphics card solder comprises:

[0032] Adding a corresponding channel identifier to the texture feature vector and the thermodynamic feature vector;

[0033] Connecting the texture feature vector and the thermodynamic feature vector with the channel identifier in the channel direction to obtain an initial fusion tensor of the graphics card solder;

[0034] Flattening the spatial dimension of the initial fusion tensor into a column vector, and splicing the column vector with the channel dimension of the initial fusion tensor as a row index to obtain the fusion feature matrix of the graphics card solder.

[0035] In a preferred embodiment, the positioning the coordinates of the false solder in the solder region based on the fusion feature matrix and a standard solder feature library comprises:

[0036] Extracting the cosine similarity of the fusion feature matrix and the standard feature library;

[0037] detecting an abnormal region with the cosine similarity lower than a preset threshold, eliminating a concave region in the abnormal region, and connecting adjacent abnormal regions to obtain an expansion region of the GPU solder joint;

[0038] taking a center coordinate of a circumscribed rectangle of the expansion region as a coordinate of a fake solder joint in the solder joint region.

[0039] In a preferred embodiment, the generating of the solder joint fake solder defect report of the GPU based on the coordinate and the texture fracture index of the fake solder joint comprises:

[0040] converting the coordinate into a physical position of the fake solder joint in a coordinate system of a circuit board;

[0041] calibrating a profile fracture amount of the GPU solder joint based on the physical position to obtain a texture fracture index of the fake solder joint;

[0042] associating the texture fracture index with a standard defect level to obtain the solder joint fake solder defect report of the GPU.

[0043] In a preferred embodiment, the calibrating of the profile fracture amount of the GPU solder joint based on the physical position to obtain the texture fracture index of the fake solder joint comprises:

[0044] obtaining an edge line of the fake solder joint through connection of multi-level gradient filtering and an adaptive threshold;

[0045] identifying a continuity interruption section of the edge line, and accumulating a geometric distance of pixels between end points of the interruption section to obtain a fracture length value;

[0046] taking a ratio of the fracture length and a solder joint perimeter as the texture fracture index of the fake solder joint.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1. By comparing the gray scale normalized visible light image and the noise suppressed infrared thermal distribution map position by position, combining the accurate alignment of texture key points and thermal gradient vector field, and effectively filtering abnormal curvature features to determine the effective solder joint region, the accuracy of the extraction of the solder joint surface texture features and thermodynamic features is improved. At the same time, by dynamically allocating feature weights to form an associated feature weight matrix, the correlation between different features is enhanced, and the extracted solder joint features are more in line with the actual state, providing a more reliable basis for subsequent detection.

[0049] 2. In the feature vector extraction, an adaptive offset sampling is adopted to accurately obtain the texture and thermodynamic feature vectors of the solder joint area, the virtual solder coordinates are located by combining the cosine similarity comparison, and the defect level is calibrated through the texture fracture index, thereby improving the positioning accuracy of the virtual solder joint and the accuracy of the defect evaluation. The generated virtual solder defect report of the solder joint can clearly reflect the virtual solder position and defect degree, thereby enhancing the accuracy of the detection result and being helpful for efficiently identifying the virtual solder problem of the graphics card. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flowchart of a graphics card solder virtual solder automatic detection method based on image recognition provided by an embodiment of the present application is shown.

[0051] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are merely intended to explain the present application and not to limit the present application.

[0053] The embodiment of the present application provides a graphics card solder virtual solder automatic detection method based on image recognition. The execution subject of the graphics card solder virtual solder automatic detection method based on image recognition includes but is not limited to at least one of electronic devices such as a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the present application. In other words, the graphics card solder virtual solder automatic detection method based on image recognition can be executed by software or hardware installed in a terminal device or a server device. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster and the like. The server can be a stand-alone server, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and big data and artificial intelligence platforms and the like basic cloud computing services.

[0054] Referring to Figure 1 A flowchart of a graphics card solder virtual solder automatic detection method based on image recognition provided by an embodiment of the present application is shown. In the embodiment, the graphics card solder virtual solder automatic detection method based on image recognition includes:

[0055] In the embodiment of the present application, when the visible light image of the graphics card after gray scale normalization and the infrared thermal distribution map of the graphics card after noise suppression are compared position by position, the surface texture features and thermodynamic features of the solder joints in the graphics card are obtained, and the surface texture features and thermodynamic features are specifically used for:

[0056] extracting texture key points in the visible light image and thermal gradient vector field of each pixel point in the infrared thermal distribution map;

[0057] filtering abnormal curvature features in the thermal gradient vector field to obtain an effective solder region of the graphics card;

[0058] aligning the texture key points and the thermal gradient vector field according to the effective solder region to obtain surface texture features and thermodynamic features of the solder of the graphics card.

[0059] Specifically, the texture key points in the visible light image are extracted to realize preliminary classification and identification by using the uniqueness of different object surface textures, and to remain stable under interference such as light change, scale scaling, slight rotation, etc., to provide reliable anchor points for subsequent processing, and to focus on local significant texture area to retain microstructure details of the object surface; the thermal gradient vector field of each pixel point in the infrared thermal distribution map is extracted to quantify the spatial features of the thermal distribution from the dynamic dimension of temperature, and the size of the vector can intuitively reflect the severity of temperature change, which can quickly locate abnormal heat areas, and the direction indicates the dominant transmission direction of heat flow to analyze the heat source diffusion path and heat conduction efficiency, and the distribution mode of the vector field can also distinguish stable heat field and dynamic heat field to provide quantitative basis for thermal state analysis.

[0060] Specifically, by eliminating the distorted curvature information in the vector field caused by noise, interference or abnormal thermal characteristics of non-solder regions, the thermal gradient distribution law specific to the solder region is accurately focused on, false features caused by uneven heat dissipation, external heat source interference and other factors are excluded, and the interference of invalid data on subsequent analysis is reduced, at the same time, the continuity and consistency features of the thermal gradient vector of the solder region are strengthened to provide clear thermal feature basis for accurately defining the spatial range of the effective solder, and to ensure that the extracted solder region has high purity and integrity.

[0061] Specifically, by aligning the texture key points and the thermal gradient vector field, the surface texture features and thermodynamic features of the graphics card solder are obtained, by matching the region range, the texture information from the visible light image and the thermal information of the infrared thermal distribution map are accurately associated to the same solder region, eliminating the spatial misplacement caused by the differences in collection angle and scale between the two modal data, making the structure details of the solder surface reflected by the texture key points and the temperature change law embodied by the thermal gradient vector field form a one-to-one spatial mapping relationship, realizing the fusion expression of double features on a single solder region, and providing complete and related basic data for subsequent feature analysis.

[0062] Further, the visible light texture key points provide static structural information such as the geometric shape and surface details of the target, which are used for positioning, contour extraction and identity recognition of the target, and the infrared thermal gradient vector field supplements the temperature variation characteristics of the target, which is used for analyzing the functional state of the target, especially in insufficient light, shielding and other visible light failure scenarios, and can independently provide a judgment basis in the thermal dimension; the two work together through multi-modal feature fusion to overcome the limitations of a single modality and improve the reliability of the scheme in complex scenarios.

[0063] Further, to provide accurate regional positioning basis for subsequent, through filtering of effective welding point area obtained by abnormal curvature characteristics, the analysis range is focused on the key connection parts of the display card, which provides a clear target area for subsequent welding point state evaluation, avoids misjudgment caused by invalid area interference, and at the same time, the area serves as a related node of thermal and structural characteristics, which can connect the extraction results of the texture key points and the thermal gradient vector field in the early stage, and provide specific action objects for multi-modal feature fusion analysis.

[0064] Further, the texture key points and the thermal gradient vector field extracted in the early stage are spatially aligned, so that the scheme can obtain the structural health and thermal stability of the welding point at the same time, break the limitations of single modal feature analysis, and provide more comprehensive judgment basis for the scheme, and the aligned features can be directly used as the input of subsequent classification, recognition or fault diagnosis model, thereby improving the accuracy and reliability of the scheme for welding point state evaluation.

[0065] In summary, visible light is vulnerable to environmental lighting in special scenarios, and infrared lacks texture details, while the extraction of texture key points and thermal gradient vector field realizes two-dimensional feature coverage of structural texture and temperature dynamics, provides complete data basis for subsequent fusion analysis, and improves feature expression capability; it provides a reusable technical paradigm for cross-modal feature fusion, establishes a double-modal feature extraction standard by clearly defining the difference between feature dimensions, and provides a theoretical reference for feature fusion in more complex multi-modal scenarios.

[0066] In summary, the abnormal feature filtering improves the signal-to-noise ratio of the thermal gradient vector field data, solves the problem of multiple thermal distribution interference in complex electronic equipment and the problem that effective features are easily hidden, and perfects the fine processing method from raw thermal data to target area, thereby providing technical support for high-precision thermal feature analysis, and the accurate extraction of the effective welding point area can improve the efficiency and accuracy of welding point quality detection, discover potential connection faults in time, provide accurate guidance for fault troubleshooting, and reduce costs and time.

[0067] In general, precise alignment of texture and thermal features is achieved through effective area constraints, providing feasible technical support for multi-feature collaborative analysis of solder joints, improving the utilization efficiency and analysis depth of feature data; at the same time, in the graphics card maintenance link, abnormal correlation analysis of dual features can be used to quickly locate faulty solder joints, shortening the troubleshooting time.

[0068] In an embodiment of the present invention, filtering the abnormal curvature features in the thermal gradient vector field to obtain the effective solder joint area of ​​the graphics card is specifically used to:

[0069] Taking the divergence value point exceeding the preset curvature threshold in the thermal gradient vector field as the extreme curvature point of the isothermal line of the graphics card solder joint;

[0070] The area where the extreme point of curvature of the isothermal line overlaps with the key point of the texture is used as the effective soldering point area of ​​the graphics card.

[0071] Specifically, by setting a scientific curvature threshold to screen the divergence value points in the thermal gradient vector field, the key points where the curvature changes significantly in the solder joint isotherms can be accurately identified. These points can intuitively reflect the mutation characteristics of the temperature distribution in the solder joint area and are the iconic locations where heat flow is concentrated or turbulent. By extracting such extreme points, the flat area information in the thermal gradient vector field can be stripped away, focusing on thermal anomaly feature points with diagnostic value, providing clear target objects for subsequent analysis of the thermal stability of the solder joint.

[0072] Specifically, the extreme points of the isothermal curvature reflect the significant changes in the thermal properties of the solder joint, while the texture key points mark the key texture features of the solder joint surface structure. The overlapping area of ​​the two means the area with both significant thermal characteristics and unique structural characteristics. This screening process can effectively exclude interfering areas with only a single modal feature, ensuring that the extracted effective solder joint area has clear solder joint characteristics in both structural and thermal properties, providing a highly reliable target area for subsequent analysis.

[0073] Furthermore, it provides a key basis for identifying thermal anomalies for subsequent analysis. By determining the extreme points of isothermal curvature, the thermal characteristic analysis of the solder joint area can be focused from the overall distribution to the local mutation points. The solution can compare the distribution density and position characteristics of these points with the preset standards, quickly lock the solder joint areas that may have quality problems, and provide accurate thermal anomaly positioning reference for subsequent fusion analysis of texture features and thermal features, thereby enhancing the solution's ability to identify solder joint failures at an early stage.

[0074] Furthermore, by locking the overlapping area of ​​the isothermal curvature extreme point and the texture key point, a solder joint area definition with both structural integrity and thermal validity is provided for subsequent analysis, allowing the solution to focus on the solder joint area that is truly functional, avoiding analysis deviations caused by vague area definition. At the same time, this area, as the intersection of structural and thermal characteristics, provides a precise spatial range for subsequent fusion analysis of the correlation between the surface texture defects and thermal anomalies of the solder joint, thereby improving the accuracy of the solution in judging solder joint quality.

[0075] In general, a quantitative screening standard for effective feature points in the thermal gradient vector field was established, and the automatic and accurate extraction of thermal anomaly points was achieved by presetting the curvature threshold, which solved the problem of large data volume and difficulty in separating effective information in the thermal gradient vector field, and improved the efficiency and pertinence of thermal feature analysis.

[0076] In general, verifying the spatial overlap of thermal and structural features solves the problem of single modal feature positioning being susceptible to interference, improves the accuracy and robustness of effective area extraction, and improves the technology for defining key areas of complex electronic components. At the same time, accurate definition of effective solder joint areas can improve the efficiency of judging solder joint quality, reduce false detections and missed detections, reduce production costs, and quickly lock the positions of solder joints that require key inspection.

[0077] In an embodiment of the present invention, the surface texture features and the thermodynamic features are combined to obtain a correlation feature weight matrix of the graphics card solder joints, specifically for:

[0078] splicing the surface texture features and the thermodynamic features along the feature dimension to obtain a high-dimensional tensor of the graphics card solder joint;

[0079] Suppressing high-frequency noise in the high-dimensional tensor;

[0080] Obtaining mask coefficients of the surface texture feature and the thermodynamic feature according to the dynamic distribution of the suppressed high-frequency noise and the weights of the surface texture feature and the thermodynamic feature;

[0081] The mask coefficients are stacked with the spatial feature vectors of the high-dimensional tensor according to pixel positions to obtain a correlation feature weight matrix of the graphics card solder joints.

[0082] Specifically, the two originally scattered modal features are structured and integrated, and a unified data carrier containing the surface microstructure information and thermal dynamic information of the solder joint is formed through the expansion of the feature dimension. This splicing method retains the original attributes and details of each feature, while establishing a correlation between different feature dimensions, avoiding the problem of isolated information in a single feature dimension, and enabling the high-dimensional tensor to comprehensively and compactly present the full picture of the multi-dimensional features of the solder joint, providing a structured, high-information data foundation for subsequent in-depth analysis.

[0083] Specifically, by suppressing high-frequency noise in the high-dimensional tensor, interference information can be effectively eliminated, making the original data of surface texture features and thermodynamic features purer, and reducing the adverse effects of noise on feature analysis. At the same time, based on the dynamic allocation of the weights of the two features and the mask coefficients after suppressing high-frequency noise, the weight allocation can be more in line with the actual situation of the data, avoiding the problem of unreasonable feature emphasis caused by fixed weights.

[0084] Specifically, the mask coefficients and the spatial feature vectors of the high-dimensional tensor are stacked according to the pixel point positions, which can organically integrate the mask coefficients reflecting the importance of surface texture features and thermodynamic features with the spatial feature vectors embodying spatial position information, forming a matrix containing feature weights and spatial position correlation. The precise matching of weight information and spatial information is realized, and the feature weight of each pixel point can correspond to its spatial position.

[0085] Further, by constructing a high-dimensional tensor to organically integrate surface texture features and thermodynamic features, the scheme can simultaneously utilize the structural features and thermal features of the solder joint for joint analysis, breaking through the limitations of single feature analysis, supporting subsequent feature mining, pattern recognition or fault classification, etc., providing data support for the scheme to output accurate evaluation results, and improving the analysis ability of the scheme for complex solder joint states.

[0086] Further, the optimized feature weights and mask coefficients are provided for subsequent analysis, decision-making and other links based on surface texture features and thermodynamic features, ensuring that the subsequent links can more accurately utilize effective features, and improving the reliability and effectiveness of the entire scheme.

[0087] Further, the mask coefficients obtained in the early stage are integrated with the spatial feature vectors into an associated feature weight matrix, which not only retains the weight information of different features, but also maintains the spatial distribution relationship of the pixel points, so that subsequent feature extraction, anomaly detection and other operations of the graphics card solder joint can be based on complete spatial-weight correlation data, ensuring the coherence of the analysis process and the integrity of the data.

[0088] In summary, the high-dimensional tensor form preserves the integrity and relevance of the features, provides an adaptive input form for deep learning, etc., promotes the technical leap of solder joint feature analysis from single dimension to multi-dimension and deep level, solves the problem of difficult collaborative analysis of multi-source features, and at the same time, the high-dimensional tensor can provide more comprehensive feature basis for the automatic detection system, improving the accuracy of solder joint quality grading.

[0089] In general, the suppression of high-frequency noise ensures the quality of feature data, lays the foundation for the accurate extraction and analysis of surface texture and thermodynamic characteristics, and helps to unearth more real and valuable information; at the same time, dynamically allocating weights and obtaining mask coefficients enables flexible adjustment of the importance of different features, enabling the scheme to better adapt to dynamic changes in data, and improving the adaptability and accuracy of the scheme when processing complex, variable, high-dimensional tensor data.

[0090] Overall, this provides accurate data support for identifying key solder joint areas and assessing solder joint quality. This avoids the disconnect between weight information and spatial information, allowing subsequent analysis to more specifically focus on important features and key locations, thereby improving the accuracy and efficiency of graphics card solder joint inspection and evaluation.

[0091] In an embodiment of the present invention, the extraction of the texture feature vector and the thermodynamic feature vector of the solder joint area in the associated feature weight matrix based on the solder joint area of ​​the graphics card is specifically used to:

[0092] Driving the sampling points of the associated feature weight matrix to adaptively shift toward the solder point deformation direction to obtain a sampling feature map of the graphics card solder point;

[0093] Taking the sampling coordinate points after affine transformation as a reference, extracting feature values ​​within the contour boundary of the graphics card solder point from the sampling feature map, and using the feature values ​​as the texture feature vector and thermodynamic feature vector set of the graphics card solder point;

[0094] A feature vector whose thermodynamic confidence weight is greater than a threshold value is selected from the thermodynamic feature vector set as the thermodynamic feature vector of the graphics card solder joint.

[0095] Specifically, by adaptively shifting the sampling points of the associated feature weight matrix toward the weld deformation direction, the sampling points can be more accurately focused on the key areas where the weld deformation occurs. This shift is not random, but dynamically adjusted based on the actual weld deformation direction. This allows the sampling points to follow the deformed area, capturing information that better reflects the weld deformation characteristics. The resulting sampling feature map can more accurately represent the state and details of the weld deformation.

[0096] Specifically, the interference of irrelevant information outside the contour is eliminated to ensure that the extracted eigenvalues ​​are completely derived from the solder joint itself. At the same time, the eigenvalues ​​are clearly distinguished into texture eigenvectors and thermodynamic eigenvector sets, achieving clear separation and precise extraction of different types of features, providing pure and clearly classified data for subsequent feature analysis.

[0097] Specifically, eigenvectors with low confidence and insufficient reliability are eliminated to ensure that the finally selected thermodynamic eigenvectors have high credibility and representativeness, avoid interference of low-quality eigenvectors in subsequent analysis, and thus improve the validity and accuracy of the characteristic data.

[0098] Furthermore, the adaptive offset processing of the sampling points combines the abstract weight association information with the actual deformation characteristics of the solder joints, providing intuitive and targeted feature image data for subsequent analysis and evaluation of the solder joint quality, playing a role in data conversion and information deepening, ensuring that the solution can transition from abstract data to concrete feature expression.

[0099] Furthermore, the coordinate reference after affine transformation is accurately extracted, and the key area information focused on by the adaptive offset of the sampling points is continued, which is converted into a structured texture and thermodynamic feature vector set, providing directly usable feature data for subsequent defect detection of graphics card solder joints and other links.

[0100] Furthermore, the feature vector is purified through threshold screening, which provides high-quality thermodynamic feature data for the subsequent comprehensive analysis by combining the texture feature vector with the thermodynamic feature vector, thereby optimizing data input and improving the efficiency of subsequent analysis.

[0101] In general, shifting the sampling points in the direction of deformation can highlight the characteristics of the solder joint deformation area, avoiding the omission of effective information caused by fixing the sampling points in non-critical areas, and improving the targetedness and effectiveness of feature extraction. At the same time, the adaptive offset feature allows the analysis to adapt to the different deformation conditions of different solder joints, enhancing the solution's adaptability to complex and variable solder joint conditions. The generated sampling feature map can more accurately determine whether the solder joint deformation is within a reasonable range and whether there are potential defects, providing a reliable basis for quality inspection and performance assurance of graphics card solder joints.

[0102] In general, extracting eigenvalues ​​within the contour based on the coordinate points after affine transformation ensures the spatial accuracy and regional specificity of the features, avoids feature misextraction due to coordinate deviation, improves the reliability of feature data, and distinguishes the eigenvalues ​​into texture and thermodynamic feature vector sets, so that subsequent analysis can conduct special evaluations of the appearance texture state and thermal performance characteristics of the solder joint, respectively, which helps to judge the quality of the solder joint more comprehensively and meticulously.

[0103] In general, the correlation between the characteristic data and the actual thermal state of the solder joint is enhanced, providing a reliable basis for accurately evaluating key indicators such as the thermal stability and heat dissipation performance of the solder joint. At the same time, the number of characteristic vectors is reduced through threshold screening, which reduces the complexity of subsequent data processing and improves the operating efficiency of the solution.

[0104] In an embodiment of the present invention, the calculation formula of the thermodynamic confidence weight is:

[0105]

[0106] Where, is the thermodynamic confidence weight, is the change in thermal conductivity, is the temperature gradient mode length, is the specific heat capacity, is the material density.

[0107] Specifically, It is the inherent property of a material's ability to conduct heat. is the change in thermal conductivity when the material state changes, and the temperature gradient is a vector describing the rate of change of temperature in space, the modulus For its scalar, The amount of heat required to raise the temperature of a unit mass of a substance by 1°C. It is the mass of a substance per unit volume.

[0108] Furthermore, Reflects the relationship between thermal properties and the driving force of heat transfer: Reflects the change of thermal conductivity, It reflects the degree of unevenness of the temperature field, and the product of the two is the coupling intensity of the change in thermal properties and the activity of heat transfer. is the volumetric specific heat capacity, which reflects the material's ability to store heat: The larger it is, the stronger the material's ability to store heat is, and the more obvious its buffering effect on thermal processes is.

[0109] In general, and Positively correlated. When the thermal conductivity of the material fluctuates more violently, Increase; When the thermal conductivity is more stable, Decrease. and The greater the temperature gradient, the stronger the driving force for heat transfer. The larger the temperature, the smaller the temperature gradient, that is, the more uniform the temperature field. The smaller. and The larger the specific heat capacity, the stronger the heat storage capacity of the material. The smaller the specific heat capacity, the weaker the heat storage capacity. The bigger. and The greater the density, the greater the mass per unit volume, which indirectly enhances the heat storage capacity. The smaller the density, the smaller the mass per unit volume. So The changing trend increases with the increase of thermal conductivity change and temperature gradient modulus, and decreases with the increase of specific heat capacity and density.

[0110] In an embodiment of the present invention, the splicing of the texture feature vector and the thermodynamic feature vector according to the channel dimension to obtain the fusion feature matrix of the graphics card solder joint is specifically used to:

[0111] adding corresponding channel identifiers to the texture feature vector and the thermodynamic feature vector;

[0112] Connecting the texture feature vector with the channel identifier and the thermodynamic feature vector along the channel direction to obtain an initial fusion tensor of the graphics card solder joint;

[0113] The spatial dimension of the initial fusion tensor is flattened into a column vector, and the column vector is spliced ​​using the channel dimension of the initial fusion tensor as a row index to obtain a fusion feature matrix of the graphics card solder joint.

[0114] Specifically, corresponding channel identifiers are added to texture feature vectors and thermodynamic feature vectors, enabling clear distinction and identification of the two different types of feature vectors. Dedicated channel identifiers accurately define the category to which each feature vector belongs, preventing feature confusion in subsequent processing, ensuring traceability of feature vector attributes, and providing an intuitive basis for distinguishing features individually.

[0115] Specifically, the information of the two types of features is retained, and they are formed into a unified whole through the association of channel dimensions, so that the originally scattered texture and thermodynamic features can be associated with each other in the same data structure, providing a complete data carrier for subsequent comprehensive analysis of the features.

[0116] Specifically, the tensor form originally containing spatial and channel information is converted into a more concise and regular matrix form, which not only retains the feature details of the spatial dimension and the category distinction of the channel dimension, but also allows the data to present structured features with clear rows and columns, which facilitates subsequent matrix operations and data analysis and processing.

[0117] Furthermore, by adding channel identifiers to mark the two types of features, a structured basis is provided for their subsequent fusion, analysis or input into the model, ensuring that different types of features can be accurately identified and processed in subsequent links.

[0118] Furthermore, the two types of features are transformed from an independent state to a fused state through the connection operation, which provides basic data for the subsequent processing of the fused tensor. This is an important transition from feature separation processing to feature comprehensive analysis.

[0119] Furthermore, through structural reorganization, it is converted into a fused feature matrix, which provides an adaptive data format for subsequent matrix operation-based feature extraction, model training or quality assessment, ensuring the smooth connection of the data processing flow and enabling the feature information fused in the early stage to participate in the subsequent analysis in a more easy-to-operate form.

[0120] In general, the addition of channel identifiers makes the boundary between texture features and thermodynamic features clearer, making it easier to focus on a certain type of feature and conduct comparative studies between the two types of features in subsequent analysis, thereby improving the flexibility and accuracy of feature processing. At the same time, this standardized identification method lays the foundation for the efficient management and application of feature vectors, can reduce errors caused by feature confusion, and plays an important supporting role in the accurate analysis and application of graphics card solder joint features.

[0121] In general, the formation of the initial fusion tensor breaks down the barriers between texture features and thermodynamic features, allowing the information of the two to work synergistically, helping to uncover related information that is difficult to discover in single feature analysis, and improving the comprehensiveness of the description of graphics card solder joint features. At the same time, the connection along the channel direction retains the channel identification information of the features, so that in subsequent analysis, we can not only use the fused overall features, but also trace back to the original texture or thermodynamic features through the channel.

[0122] In general, the complexity of subsequent data processing is reduced, allowing the matrix to be directly applied to feature analysis, improving the efficiency of data processing. At the same time, the spatial and channel information in the initial fusion tensor is fully preserved, ensuring the integrity and relevance of the features, allowing subsequent analysis to be performed based on comprehensive feature data.

[0123] In an embodiment of the present invention, the coordinates of the cold solder joints in the solder joint area are located based on the fusion feature matrix and the standard solder joint feature library, specifically for:

[0124] Extracting the cosine similarity between the fused feature matrix and the standard feature library;

[0125] Detecting abnormal regions where the cosine similarity is lower than a preset threshold, eliminating concave regions in the abnormal regions, and connecting adjacent abnormal regions to obtain an expanded region of the graphics card solder joint;

[0126] The center coordinates of the circumscribed rectangle of the expansion area are used as the coordinates of the cold solder joints in the solder joint area.

[0127] Specifically, the cosine similarity can effectively reflect the similarity of two vectors in the direction, and is not affected by the length of the vector. Here, the degree of fit between the analyzed graphics card solder joint fusion features and the standard feature library can be accurately measured, and the similarity value obtained can be used as a direct quantitative index to determine whether the solder joint features meet the standard.

[0128] Specifically, first, accurately locate the abnormal area with a cosine similarity lower than the preset threshold. These areas are often where the graphics card solder joint may have defects. Then, eliminate the concave areas in the abnormal area to avoid interference from the details of the concave part in the overall abnormal area judgment, making the area shape more regular. Next, connect adjacent abnormal areas to integrate scattered small abnormal areas into complete large areas, and finally obtain the inflation area.

[0129] Specifically, by taking the center coordinates of the circumscribed rectangle of the inflation area as the virtual solder joint coordinates, the irregular shape of the inflation area can be ignored, and a unified coordinate point is used to lock the virtual solder joint position, avoiding positioning ambiguity caused by complex area shape, and ensuring that each virtual solder joint area has a unique and easily identifiable position reference.

[0130] Further, with the help of cosine similarity calculation with the standard feature library, abstract feature data is converted into similarity results with clear judgment significance, providing a key basis for subsequent judgment of graphics card solder joint quality, defect identification, etc.

[0131] Further, by detecting and processing the abnormal area, abstract similarity data is converted into concrete abnormal area shape, providing a clear area basis for subsequent accurate identification of graphics card solder joint defects, severity assessment, etc., and playing a transitional role from abnormal judgment to defect visualization.

[0132] Further, the inflation area is converted into specific coordinate information, providing direct position guidance for subsequent virtual solder joint confirmation, repair operations, etc., and is a key link connecting abnormal area analysis and actual production application.

[0133] Overall, the extraction of cosine similarity achieves objective and quantitative comparison of graphics card solder joint features and standard features, avoiding subjective judgment bias and improving the accuracy and consistency of quality assessment. At the same time, this quantitative method can sensitively capture subtle differences between features, helping to timely discover potential defects or deviations from the standard in the solder joint.

[0134] Overall, detecting and processing the abnormal area can accurately lock the problem area of the graphics card solder joint, and eliminating the concave area and connecting adjacent areas can avoid missed detection or misjudgment, improving the accuracy and integrity of abnormal area identification. At the same time, the inflation area obtained directly shows the range and shape of the solder joint anomaly, providing a reliable reference for analyzing the causes of defects and developing repair solutions.

[0135] In general, the unified coordinate positioning method ensures the standardization and traceability of the test results, facilitates statistical analysis of cold solder joints, and provides data support for reducing the incidence of cold solder joints. It upgrades cold solder joint detection from qualitative judgment to quantitative positioning.

[0136] In an embodiment of the present invention, when generating the solder joint defect report of the graphics card according to the coordinates and the texture fracture index of the solder joint, it is specifically used to:

[0137] Converting the coordinates into the physical position of the cold solder joint in the circuit board coordinate system;

[0138] Calibrate the contour fracture amount of the graphics card solder joint based on the physical position to obtain the texture fracture index of the virtual solder joint;

[0139] The texture fracture index is associated with a standard defect level to obtain a solder joint cold soldering defect report of the graphics card.

[0140] Specifically, digital coordinates are converted into physical locations directly corresponding to the circuit board. This eliminates the differences between different coordinate systems, allowing the detected coordinate information to correspond to the actual structure and component layout of the circuit board, providing a direct basis for finding the exact location of the cold solder joint on the physical circuit board.

[0141] Specifically, through physical position calibration, the calculation error of contour fracture caused by position deviation is eliminated, so that the obtained fracture amount is more in line with the actual situation; and the texture fracture index converts the abstract contour fracture information into a quantifiable indicator, which intuitively reflects the degree of texture fracture of the cold solder joint and provides a specific basis for judging the severity of the cold solder joint.

[0142] Specifically, by correlating the texture fracture index with standard defect levels, we can transform quantified texture fracture data into defect levels with clear evaluation criteria, ultimately forming a systematic and standardized report on solder joint defects in graphics cards. This transformation from specific data to conclusive reports provides a unified standard for evaluating the severity of solder joint defects. The report contains both quantitative evidence and clear grading, making it easy to intuitively understand and use.

[0143] Furthermore, it is mapped to the physical space of the circuit board through coordinate system conversion, providing precise position guidance for subsequent automation and other practical operations, completing the key transition from data information to physical location, and ensuring the practicality and operability of the entire solution.

[0144] Furthermore, by calibrating the contour fracture amount and calculating the texture fracture index, the physical position information is combined with the texture characteristics of the solder joint, providing feature data for subsequent quality rating of the cold solder joints and repair priority determination.

[0145] Furthermore, the texture fracture index obtained in the early stage is correlated with the standard defect level to comprehensively evaluate the solder joint condition of the graphics card and generate a report, providing the final decision-making basis for subsequent links.

[0146] In general, it can accurately find the cold solder joints on the circuit board, avoiding problems such as finding the wrong position due to coordinate mismatch, improving maintenance efficiency, and at the same time unifying the physical position information under the circuit board coordinate system, providing a spatial coordinate basis for tracing the cause of the cold solder joint.

[0147] In general, the calibration based on physical position ensures the accuracy of the contour fracture quantity, and the introduction of the texture fracture index realizes the quantitative evaluation of the texture characteristics of the cold weld, which upgrades the judgment of the severity of the cold weld from qualitative description to quantitative analysis, improving the objectivity and accuracy of the evaluation. At the same time, the index provides a more detailed reference basis for subsequent analysis, which helps to formulate differentiated repair strategies according to the size of the fracture index.

[0148] In general, the deviation in information transmission is reduced. At the same time, the texture fracture index and standard defect level associated with the report provide detailed reference data for tracing the root cause of the cold solder joint problem, which helps to reduce the incidence of cold solder joints. The clear defect level also provides a basis for determining maintenance priorities and improves the utilization efficiency of maintenance resources.

[0149] In an embodiment of the present invention, when calibrating the contour fracture amount of the graphics card solder joint based on the physical position to obtain the texture fracture index of the cold solder joint, it is specifically used to:

[0150] The edge line of the virtual solder joint is obtained by connecting multi-level gradient filtering with an adaptive threshold;

[0151] Identifying the continuity interruption segment of the edge line, and accumulating the geometric distance of pixels between the endpoints of the interruption segment to obtain the break length value;

[0152] The ratio of the fracture length to the perimeter of the solder joint is used as the texture fracture index of the cold solder joint.

[0153] Specifically, multi-level gradient filtering is used to enhance the difference between the edge of the cold solder joint and the surrounding area, highlighting the edge features. Then, the adaptive threshold is combined to connect the edge fragments, and finally a complete and continuous cold solder joint edge line is formed, which effectively filters out noise interference and accurately captures edge details. It solves the problem of edge breakage and blur under single filtering or fixed threshold, and makes the outline boundary of the cold solder joint clearly discernible.

[0154] Specifically, the discontinuous features of the edge line are converted into specific numerical indicators, which clearly reflect the degree and range of the edge fracture of the cold solder joint, avoid the vague description of the edge integrity, and provide an intuitive and comparable quantitative basis for judging the edge damage.

[0155] Specifically, the problem of incomparable absolute values ​​of fracture lengths of different solder joints due to size differences is eliminated, so that the relative severity of texture fracture can be measured by a unified index regardless of the size of the solder joint.

[0156] Furthermore, the extracted edge lines further depict the geometric shape of the cold solder joints, providing accurate edge basis for subsequent contour parameter calculation, defect morphology classification, etc., and playing a transitional role from regional positioning to morphology quantification.

[0157] Furthermore, the morphological features of the edge lines are further converted into quantifiable parameters, providing important feature data for subsequent evaluation of the severity of cold solder defects and classification of defects, thereby improving the capture of multi-dimensional features of cold solder joints.

[0158] Furthermore, the specific information of edge fractures is converted into indicators with clear evaluation significance, which provides a key quantitative basis for the subsequent correlation of standard defect levels and generation of defect reports, and plays a role in the transformation from raw data to evaluation indicators.

[0159] In general, edge features can be used to analyze solder joints for deformation, incompleteness, and other issues, improving the precision of defect identification. Furthermore, the combination of multi-level gradient filtering and adaptive thresholding allows the process to adapt to solder joint images under varying lighting and material conditions, enhancing the solution's adaptability to complex scenarios. Accurate edge line extraction provides the foundation for subsequent calculations of solder joint size and shape parameters, enabling a more scientific assessment of the severity of cold solder joint defects and playing a crucial role in improving graphics card solder joint quality inspection systems.

[0160] In general, it is convenient to quickly distinguish the severity of defects and provide data support for formulating differentiated treatment strategies. At the same time, this value can accurately reflect the actual scale of edge fracture and help to deeply analyze the causes of cold solder joints.

[0161] In general, it objectively reflects the degree of damage to the texture integrity of the solder joint caused by the cold solder joint, avoids misjudgment due to the different sizes of the solder joints themselves, and improves the accuracy and fairness of defect assessment. At the same time, the index provides a unified standard for the horizontal comparison between different cold solder joints and the longitudinal monitoring of the same solder joint at different stages, which is convenient for statistical analysis and trend judgment of cold solder joint problems in production.

[0162] Compared with the prior art, the present invention has the following beneficial effects:

[0163] 1. By comparing the grayscale-normalized visible light image with the noise-suppressed infrared thermal map, and combining precise alignment of texture keypoints with the thermal gradient vector field, the system effectively filters out anomalous curvature features to identify valid solder joint areas, improving the accuracy of extracting both surface texture and thermodynamic features. Furthermore, by dynamically assigning feature weights to form a correlation feature weight matrix, the correlation between different features is enhanced, making the extracted solder joint features more accurate and accurate, providing a more reliable foundation for subsequent inspections.

[0164] 2. Adaptive offset sampling is used in feature vector extraction to accurately obtain the texture and thermodynamic feature vectors of the solder joint area. This is combined with cosine similarity comparison to locate the coordinates of the cold solder joint, and the defect level is calibrated using the texture fracture index. This improves the accuracy of cold solder joint location and defect assessment. The generated cold solder joint defect report clearly reflects the location and degree of the cold solder joint, enhancing the accuracy of the inspection results and helping to efficiently identify cold solder joint issues in graphics cards.

[0165] In the several embodiments provided by the present invention, it should be understood that the disclosed methods can be implemented in other ways.

[0166] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0167] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence refers to the theories, methods, and technologies that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for automatically detecting solder joint defects in graphics cards based on image recognition, characterized in that: The method comprises: S1. Comparing the visible light image of the graphics card after grayscale normalization and the infrared thermal distribution map of the graphics card after noise suppression position by position, to obtain the surface texture characteristics and thermodynamic characteristics of the solder joints in the graphics card; S2. combining the surface texture features and the thermodynamic features to obtain a correlation feature weight matrix of the graphics card solder joints; S3. Extracting, based on the solder joint area of ​​the graphics card, a texture feature vector and a thermodynamic feature vector of the solder joint area in the associated feature weight matrix; S4. Concatenate the texture feature vector and the thermodynamic feature vector according to the channel dimension to obtain a fusion feature matrix of the graphics card solder joint; S5. Locating the coordinates of the cold solder joints in the solder joint area based on the fused feature matrix and the standard solder joint feature library; S6. Generate a solder joint defect report of the graphics card according to the coordinates and the texture fracture index of the solder joint.

2. The method for automatically detecting solder joint defects in graphics cards based on image recognition according to claim 1, wherein: The step of comparing the visible light image of the graphics card after grayscale normalization and the infrared thermal distribution map of the graphics card after noise suppression position by position to obtain the surface texture characteristics and thermodynamic characteristics of the solder joints in the graphics card includes: Extracting the texture key points in the visible light image and the thermal gradient vector field of each pixel point in the infrared thermal distribution map; filtering abnormal curvature features in the thermal gradient vector field to obtain an effective solder joint area of ​​the graphics card; The texture key points are aligned with the thermal gradient vector field according to the effective solder joint area to obtain surface texture features and thermodynamic features of the solder joints in the graphics card.

3. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 2, wherein: The filtering of abnormal curvature features in the thermal gradient vector field to obtain an effective solder joint area of ​​the graphics card includes: Taking the divergence value point exceeding the preset curvature threshold in the thermal gradient vector field as the extreme curvature point of the isothermal line of the graphics card solder joint; The area where the extreme point of curvature of the isothermal line overlaps with the key point of the texture is used as the effective soldering point area of ​​the graphics card.

4. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 1, wherein: The step of combining the surface texture features and the thermodynamic features to obtain a correlation feature weight matrix of the graphics card solder joints includes: splicing the surface texture features and the thermodynamic features along the feature dimension to obtain a high-dimensional tensor of the graphics card solder joint; Suppressing high-frequency noise in the high-dimensional tensor; Obtaining mask coefficients of the surface texture feature and the thermodynamic feature according to the dynamic distribution of the suppressed high-frequency noise and the weights of the surface texture feature and the thermodynamic feature; The mask coefficients are stacked with the spatial feature vectors of the high-dimensional tensor according to pixel positions to obtain a correlation feature weight matrix of the graphics card solder joints.

5. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 1, wherein: The step of extracting the texture feature vector and the thermodynamic feature vector of the solder joint area in the associated feature weight matrix according to the solder joint area of ​​the graphics card includes: Driving the sampling points of the associated feature weight matrix to adaptively shift toward the solder point deformation direction to obtain a sampling feature map of the graphics card solder point; Taking the sampling coordinate points after affine transformation as a reference, extracting feature values ​​within the contour boundary of the graphics card solder point from the sampling feature map, and using the feature values ​​as the texture feature vector and thermodynamic feature vector set of the graphics card solder point; A feature vector whose thermodynamic confidence weight is greater than a threshold value is selected from the thermodynamic feature vector set as the thermodynamic feature vector of the graphics card solder joint.

6. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 5, wherein: The calculation formula of the thermodynamic confidence weight is: ； Where, is the thermodynamic confidence weight, is the change in thermal conductivity, is the temperature gradient mode length, is the specific heat capacity, is the material density.

7. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 1, wherein: The step of splicing the texture feature vector and the thermodynamic feature vector according to the channel dimension to obtain a fusion feature matrix of the graphics card solder joint includes: adding corresponding channel identifiers to the texture feature vector and the thermodynamic feature vector; Connecting the texture feature vector with the channel identifier and the thermodynamic feature vector along the channel direction to obtain an initial fusion tensor of the graphics card solder joint; The spatial dimension of the initial fusion tensor is flattened into a column vector, and the column vector is spliced ​​using the channel dimension of the initial fusion tensor as a row index to obtain a fusion feature matrix of the graphics card solder joint.

8. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 1, wherein: The locating the coordinates of the cold solder joints in the solder joint area based on the fusion feature matrix and the standard solder joint feature library includes: Extracting the cosine similarity between the fused feature matrix and the standard feature library; Detecting abnormal regions where the cosine similarity is lower than a preset threshold, eliminating concave regions in the abnormal regions, and connecting adjacent abnormal regions to obtain an expanded region of the graphics card solder joint; The center coordinates of the circumscribed rectangle of the expansion area are used as the coordinates of the cold solder joints in the solder joint area.

9. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 1, wherein: Generating a solder joint defect report of the graphics card according to the coordinates and the texture fracture index of the solder joint includes: Converting the coordinates into the physical position of the cold solder joint in the circuit board coordinate system; Calibrate the contour fracture amount of the graphics card solder joint based on the physical position to obtain the texture fracture index of the virtual solder joint; The texture fracture index is associated with a standard defect level to obtain a solder joint cold soldering defect report of the graphics card.

10. The method for automatically detecting solder joint defects in graphics cards based on image recognition as claimed in claim 9, wherein: The step of calibrating the contour fracture amount of the graphics card solder joint based on the physical position to obtain the texture fracture index of the cold solder joint includes: The edge line of the virtual solder joint is obtained by connecting multi-level gradient filtering with an adaptive threshold; Identifying the continuity interruption segment of the edge line, and accumulating the geometric distance of pixels between the endpoints of the interruption segment to obtain the break length value; The ratio of the fracture length to the perimeter of the solder joint is used as the texture fracture index of the cold solder joint.

Images