Welding quality detection method and system for high-order computing power automatic driving FPC sensing module

Abstract

The invention relates to a welding quality detection method and system for a high-order computing power automatic driving FPC sensing module, and the method comprises the following steps: carrying out the feature extraction of a multi-dimensional spectral image set, obtaining a welding feature vector of a welding region, and constructing a welding feature map based on the welding feature vector; performing dynamic feature mapping on the welding feature spectrum through a spectrum transformation network to obtain a welding feature descriptor; based on the welding feature descriptors, topological structure analysis is conducted on the welding area, and a welding association diagram is obtained; and inputting the welding association diagram into a preset region growing algorithm for quality detection and grading to obtain a welding quality evaluation report, thereby solving the problems that the traditional method generally depends on manual inspection, the efficiency is low, the influence of operator experience and subjective judgment is easy to cause, and the working efficiency is low. And the requirements of the modern manufacturing industry on automation and precision cannot be met.

Description

Technical Field

[0001] The present invention relates to the technical field of FPC sensor modules, and in particular to a welding quality detection method and system for a high-order computing power autonomous driving FPC sensor module. Background Art

[0002] With the rapid development of autonomous driving technology, the requirements for vehicle perception systems are also increasing. FPC (flexible printed circuit) sensor modules, as important components connecting sensors and control systems, play a key role in autonomous vehicles. These modules need to maintain high reliability in complex environments, so the welding quality during their manufacturing process is crucial. However, traditional welding quality inspection methods, such as visual inspection and X-ray inspection, although able to identify obvious defects, are incapable of detecting minor or hidden welding problems. In addition, traditional methods usually rely on manual inspection, which is not only inefficient but also easily affected by operator experience and subjective judgment, and cannot meet the needs of modern manufacturing for automation and precision.

[0003] Driven by the demand for high-level computing power, autonomous driving technology is evolving towards a more intelligent direction, which places higher requirements on FPC sensor modules. In order to ensure the stability and safety of autonomous driving systems, more advanced and reliable welding quality detection methods must be introduced. The development of existing multispectral imaging technology and convolutional neural networks has brought new opportunities to this field. The use of multispectral imaging can obtain more information than traditional methods, and the application of deep learning algorithms makes it possible to automatically analyze and understand these complex data. However, how to effectively combine these two advanced technologies and apply them to welding quality assessment in actual production environments is still a problem that needs to be solved.

[0004] In response to the above challenges, a welding quality detection method for a high-order computing power autonomous driving FPC sensor module based on a combination of multispectral imaging and deep learning is proposed. This method aims to overcome the limitations of existing technologies and provide an efficient, accurate and automated solution. It uses multi-angle scanning imaging to capture the multidimensional features of the welding area, and uses convolutional neural networks for feature extraction to construct a detailed welding feature map. Furthermore, dynamic feature mapping is achieved through a map transformation network, and a regional growing algorithm is used for quality inspection and grading, and finally a welding quality assessment report is generated. This method not only improves the accuracy of detection, but also greatly shortens the detection time, which helps to improve the efficiency of the entire manufacturing process and product quality. Summary of the invention

[0005] The main purpose of the present invention is to provide a welding quality detection method and system for a high-order computing power autonomous driving FPC sensor module, which solves the technical problem that traditional methods usually rely on manual inspection, which is not only inefficient but also easily affected by operator experience and subjective judgment, and cannot meet the requirements of automation and precision in modern manufacturing.

[0006] To achieve the above object, the present invention provides a welding quality detection method for a high-order computing power autonomous driving FPC sensor module, comprising the following steps: The welding area of ​​the target FPC sensor module is scanned and imaged at multiple angles by a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area; Extracting features from the multidimensional spectral image set by a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector; Performing dynamic feature mapping on the welding feature map through a map transformation network to obtain a welding feature descriptor; Performing a topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association graph; The welding association diagram is input into a preset region growing algorithm for quality inspection and grade classification to obtain a welding quality assessment report.

[0007] Furthermore, the welding area of ​​the target FPC sensor module is scanned and imaged at multiple angles by a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area, including: Performing a full spectrum scan of the welding area through a preset multi-spectral imaging system to obtain an original spectral data sequence, wherein the original spectral data sequence includes welding information of a visible spectrum, a near infrared spectrum, and an ultraviolet spectrum; Performing band separation on the original spectral data sequence to obtain a multi-band spectral characteristic graph, and performing spectral enhancement processing on the multi-band spectral characteristic graph to obtain an enhanced spectral characteristic graph; Perform three-dimensional reconstruction based on the enhanced spectral characteristic graph to obtain a welding stereoscopic spectrum graph; Based on the spectral registration network, the welding stereo spectrum image is spatially registered to obtain a registered spectral data set, and the registered spectral data set is feature fused by a spectral feature fusion algorithm to obtain a fused spectral feature image; Performing end-member analysis on the fused spectral characteristic graph by using a spectral unmixing algorithm to obtain a welding material component distribution graph, and performing geometric correction on the welding material component distribution graph to obtain a corrected component graph; wherein the welding material component distribution graph includes solder composition, substrate characteristics and interface structure information; The corrected component map is subjected to dimensionality reduction and denoising processing to obtain a multidimensional spectral image set of the welding area.

[0008] Furthermore, the feature extraction of the multidimensional spectral image set is performed by a preset convolutional neural network to obtain a welding feature vector of the welding area, and a welding feature map is constructed based on the welding feature vector, including: The spatial spectrum feature extraction is performed on the multi-dimensional spectral image set by a preset convolutional neural network to obtain a spatial spectrum feature map; wherein the spatial spectrum feature map includes spatial texture information and spectral component information of the welding area; Performing multi-scale feature fusion on the spatial spectrum feature map through a pyramid pooling module to obtain a multi-scale feature representation, and performing local binary pattern encoding on the multi-scale feature representation to obtain an LBP feature map; Performing global context modeling on the LBP feature map based on a Transformer network to obtain a global context feature vector; Performing feature compression on the global context feature vector by using a local sensitive hash coding technique to obtain a sparse feature vector, and using the sparse feature vector as a welding feature vector of the welding area; Relationship modeling is performed based on the sparse feature vector to obtain a welding feature relationship diagram, and a welding feature map is constructed based on the welding feature relationship diagram; wherein the welding feature map includes welding intensity distribution data, welding depth distribution data and welding surface flatness data.

[0009] Furthermore, the welding feature map is dynamically mapped by a map transformation network to obtain a welding feature descriptor, including: Performing multi-resolution decomposition on the welding feature atlas through a preset atlas transformation network to obtain a welding feature sub-atlas set; Based on the bidirectional long short-term memory network, the welding feature sub-atlas set is subjected to time series feature analysis to obtain a time series feature sequence, and the time series feature sequence is subjected to structured analysis through the graph attention network to obtain a welding dynamic feature graph; Performing Fourier spectrum analysis on the welding dynamic characteristic graph to obtain a frequency domain characteristic spectrum; Extract key features from the frequency domain feature spectrum using an adaptive feature selection algorithm to obtain a welding feature description vector; Performing topological encoding on the welding feature description vector based on a graph isomorphic network to obtain a structured feature encoding; The structured feature code is transformed and mapped by a graph mapping algorithm to obtain a welding feature descriptor; wherein the welding feature descriptor includes an interface bonding state, a void distribution density and a metal interconnection strength.

[0010] Further, the topological structure analysis of the welding area based on the welding feature descriptor to obtain a welding association graph includes: The welding feature descriptor is locally deconstructed by a preset welding distribution analysis algorithm to obtain a local feature distribution map of the welding area, and the local feature distribution map is subjected to feature connection analysis to obtain welding feature local connection data; wherein the welding feature local connection data includes position information, connection weight information and node type information of welding nodes; Based on the welding feature local connection data, a global network model is performed on the welding area to obtain a global topological structure diagram of the welding area, and feature mapping is performed on the global topological structure diagram to obtain an embedded topological feature vector; Inferring the dynamic adjacency relationship of the embedded topological feature vector through a topological optimization network to obtain dynamic feature association data between welding areas, and performing feature aggregation on the dynamic feature association data to obtain a welding cluster feature distribution map, which includes spatial association information of welding group structure and defect characteristics; Performing geometric structural decomposition on the welding cluster feature distribution map by using a geometric association analysis algorithm to obtain a geometric association map of the welding area, and performing high-order feature mapping on the geometric association map to obtain a welding geometric feature description map; Based on the welding geometric feature description diagram, multi-layer relationship modeling is performed on the dynamic characteristics of the welding area to obtain a welding dynamic feature diagram; The welding dynamic feature map and the geometric association map are jointly encoded by a high-order feature fusion algorithm to obtain a welding association map.

[0011] Furthermore, the welding association graph is input into a preset region growing algorithm for quality detection and grade classification to obtain a welding quality assessment report, including: The seed points of the welding association graph are located by a regional growth seed selection algorithm to obtain a welding region seed point set, and the regional growth factor is calculated for the welding region seed point set to obtain regional growth feature data; wherein the regional growth feature data includes welding connection strength, welding boundary gradient and welding region morphological characteristics; Based on the regional growth feature data, the welding region is adaptively expanded to obtain a welding region growth map, and the welding region growth map is extracted from the boundary features by a regional boundary tracking algorithm to obtain a welding region boundary feature set; Performing quality inspection and grading on the welding area boundary feature set according to a preset quality assessment criterion to obtain a welding quality grade distribution map; Based on the deep feature analysis network, the welding quality grade distribution map is subjected to deep feature analysis to obtain a welding quality feature vector, and the welding quality feature vector is subjected to hierarchical evaluation through a multi-level evaluation algorithm to obtain welding quality evaluation data; wherein the welding quality evaluation data includes a welding strength score, a welding depth score and a welding surface score; The welding quality assessment data is structured and content optimized through an assessment report generation algorithm to obtain a welding quality assessment report; wherein the welding quality assessment report includes quality grade distribution statistics, defect type analysis and process parameter recommendations.

[0012] Furthermore, the adaptively expanding the welding region based on the region growth characteristic data to obtain a welding region growth map includes: Performing dynamic threshold calculation on the regional growth feature data by using a preset regional growth threshold adaptive algorithm to obtain a regional growth threshold set, and locally optimizing the regional growth threshold based on the morphological characteristics of the welding area to obtain an optimized growth threshold set; Based on the optimized growth threshold set, the welding area is expanded in multiple directions to obtain a regional expansion direction map, and the expansion direction map is connected to mark the domain through a regional connectivity analysis algorithm to obtain a welding area connection mark map; The growth constraint calculation is performed on the welding area connected labeling graph by using a preset regional growth constraint algorithm to obtain regional growth constraint data. Based on the regional competition mechanism, multi-region competition growth is performed on the regional growth constraint data to obtain a regional competition growth map, and similar regions of the regional competition growth map are merged through a regional merging algorithm to obtain a welding region growth map.

[0013] The present invention also provides a welding quality detection system for a high-order computing power autonomous driving FPC sensor module, comprising: An imaging module is used to perform multi-angle scanning imaging on the welding area of ​​the target FPC sensor module through a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area; An extraction module, used for performing feature extraction on the multidimensional spectral image set through a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector; A mapping module, used for performing dynamic feature mapping on the welding feature map through a map transformation network to obtain a welding feature descriptor; An analysis module, configured to perform a topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association diagram; The classification module is used to input the welding association diagram into a preset region growing algorithm to perform quality grade classification and obtain a welding quality assessment report.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The welding quality detection method of the high-order computing power autonomous driving FPC sensor module provided by the present invention comprises the following steps: performing multi-angle scanning imaging on the welding area of ​​the target FPC sensor module by a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area; performing feature extraction on the multi-dimensional spectral image set by a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector; performing dynamic feature mapping on the welding feature map by a map transformation network to obtain a welding feature descriptor; performing topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association map; inputting the welding association map into a preset regional growing algorithm for quality detection and grade classification to obtain a welding quality assessment report, thereby solving the technical problem that the traditional method usually relies on manual inspection, which is not only inefficient, but also easily affected by the operator's experience and subjective judgment, and cannot meet the technical problem of the modern manufacturing industry's requirements for automation and precision, and realizes the topological structure analysis based on the welding feature descriptor, so that the detection system can deeply understand the spatial relationship between welding points and their connectivity. This in-depth structured analysis can help discover tiny defects or potential problems that are difficult to detect with traditional methods, ensuring the beneficial effect of a comprehensive assessment of welding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 1 is a schematic diagram of the steps of a welding quality detection method of a high-order computing power autonomous driving FPC sensor module in one embodiment of the present invention; Figure 2 It is a structural block diagram of a welding quality detection system of a high-order computing power autonomous driving FPC sensor module in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] like Figure 1 As shown, Figure 1 This is a schematic diagram of the steps of a welding quality detection method for a high-order computing power autonomous driving FPC sensor module in one embodiment of the present invention; In one embodiment of the present invention, a welding quality detection method for a high-order computing power autonomous driving FPC sensor module is provided, comprising the following steps: Step S1, performing multi-angle scanning imaging on the welding area of ​​the target FPC sensor module by a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area.

[0021] Specifically, in the process of realizing multi-angle scanning and imaging of the welding area of ​​the target FPC sensor module, the preset multispectral imaging system plays a vital role. This process first involves the selection and installation of suitable multispectral imaging devices that can emit and capture light in different wavelength ranges to form a series of images representing different spectral information. For example, on the production line of the FPC sensor module for autonomous driving, in order to ensure the quality of welding, engineers will set up a multispectral imaging system that can scan the welding area from multiple different angles at a specific time point. When the multispectral imaging system is started, it will gradually scan the welding area according to the predetermined angle and sequence, and each scan will generate an image reflecting the spectral characteristics of the welding area at that angle. Due to the use of multispectral technology, each image is not just an ordinary black and white or color photo, but contains the reflectivity or transmittance information of the welding area under different spectra. In this way, by integrating multiple images from different angles, a more complete and detailed multidimensional spectral image set of the welding area can be constructed. For example, in actual applications, if there are slight cracks or insufficient solder at a certain solder joint, the image under certain specific spectra will show different characteristics from normal solder joints, and these differences may be ignored in conventional visual inspections. In this process, the accuracy and stability of the system are crucial to obtaining high-quality multi-dimensional spectral images. Therefore, engineers need to carefully calibrate the multi-spectral imaging system to ensure that each scanning angle can provide clear and distortion-free images. At the same time, the influence of ambient lighting conditions must also be considered, as external light sources may interfere with the imaging effect. To solve this problem, scanning can be performed in a closed environment or filters can be used to shield unnecessary light. The resulting multi-dimensional spectral image set can not only reflect the morphological characteristics of the welding surface, but also reveal potential changes in material properties, providing a solid data foundation for subsequent feature extraction and quality assessment. This process is a key step in the entire high-level computing power autonomous driving FPC sensor module welding quality detection method, which is directly related to whether welding defects can be accurately identified, thereby ensuring the reliable operation of the sensor module in complex driving environments.

[0022] Step S2, extracting features from the multidimensional spectral image set by using a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector.

[0023] Specifically, in the process of feature extraction of the multidimensional spectral image set, the preset convolutional neural network (CNN) plays a core role. This process is to automatically identify and extract representative features from the multidimensional spectral image set of the welding area, and convert these features into welding feature vectors, and then construct a welding feature map. Specifically, after the multi-spectral imaging system completes the multi-angle scanning of the welding area of ​​the FPC sensor module, the obtained multi-dimensional spectral image set is input into the trained convolutional neural network. The convolutional neural network first processes the image data through a series of convolutional layers. In each convolutional layer, the network uses multiple small filters to perform convolution operations with different parts of the image to capture local patterns and structural information in the image. As the data is passed layer by layer, the network can gradually learn from simple to complex feature representations, such as from low-level features such as edges and textures to higher-level geometric shapes or material property features. For the application scenario of the FPC sensor module for autonomous driving, this means that CNN can identify the morphology of solder joints, the distribution of solder, and possible defects such as cracks or pores. Once the convolutional neural network has completed the feature extraction of the multidimensional spectral image set, it will output a high-dimensional feature vector, which contains all the extracted feature information. In order to make this information more intuitive and easy to analyze, the welding feature map will be constructed based on this welding feature vector. The process of constructing the map involves mapping the values ​​in the feature vector into a two-dimensional or three-dimensional space to form a map that can visually display the characteristics of the welding area. For example, in practical applications, engineers can quickly locate the problem by viewing the welding feature map and evaluate whether the welding quality meets the standard. In addition, in order to ensure that the convolutional neural network can accurately complete the feature extraction task, it must be fully trained in advance. This usually requires a large amount of labeled data as input to allow the network to learn how to distinguish between qualified and unqualified welding situations. By continuously adjusting the network parameters to minimize the prediction error, the CNN is finally able to give a reliable welding feature vector when facing a new multidimensional spectral image set. The whole process not only improves the detection efficiency, but also greatly enhances the degree of automation, providing a scientific basis for subsequent quality assessment. In this way, whether it is daily production monitoring or new product development, this method can be relied upon to ensure the consistency and reliability of the FPC sensor module welding quality, thereby ensuring the stable operation of the autonomous driving system.

[0024] Step S3, dynamically mapping the welding feature map through a map transformation network to obtain a welding feature descriptor.

[0025] Specifically, after the construction of the welding feature map is completed, the next step is to perform dynamic feature mapping on the welding feature map through the atlas transformation network. This step aims to further process and analyze the extracted feature information to obtain more abstract and descriptive welding feature descriptors. Specifically, after the convolutional neural network generates welding feature vectors and constructs corresponding welding feature maps, although these maps intuitively show the feature distribution of the welding area, they are still mainly static, image-based representations. In order to mine deeper time series changes or other dynamic characteristics from these static features, the atlas transformation network is introduced to perform dynamic feature mapping. This network can capture the possible change patterns that may occur during the welding process, such as subtle differences caused by fluctuations in welding process parameters, or the mutual influence between different welding points. In the application scenario of the FPC sensor module for autonomous driving, this means that the atlas transformation network can identify the trend of the evolution of solder joints over time, as well as the dynamic behavior of the solder during melting and cooling. For example, if a weld shrinks unevenly during cooling, this phenomenon may not be directly reflected in the static weld feature map, but through dynamic feature mapping, this nonlinear change can be converted into an easy-to-understand feature descriptor. The working principle of the graph transformation network involves a series of complex mathematical operations and model transformations. It uses the transformation mechanism in deep learning technology, such as self-attention mechanism or recurrent neural network (RNN), to simulate and learn the spatiotemporal relationship in the weld feature map. In this way, the network can automatically adjust its internal structure to adapt to the feature expression requirements of different welding situations. In practical applications, engineers can train the graph transformation network to learn how to extract the most representative dynamic features from the weld feature map and encode them into weld feature descriptors. These descriptors not only contain the spatial information in the original image, but also incorporate the change rules in the time dimension, thus providing a more comprehensive basis for welding quality assessment. In addition, in order to ensure the effectiveness and reliability of the graph transformation network, a large amount of labeled data is needed for supervised learning. This includes a set of multidimensional spectral images collected under various welding conditions and their corresponding high-quality weld feature descriptors. By continuously optimizing the network parameters, the graph transformation network can accurately perform dynamic feature mapping even when faced with new welding feature graphs. The final welding feature descriptor not only helps with subsequent topological structure analysis and quality inspection, but can also be used as feedback information to improve welding processes and increase production efficiency. In this way, during the entire manufacturing process of the high-level computing power autonomous driving FPC sensor module, the refined management and real-time monitoring of welding quality can effectively ensure the reliability and safety of the product, thereby promoting the development of autonomous driving technology.

[0026] Step S4, performing a topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association graph.

[0027] Specifically, after the construction of the welding feature descriptor is completed, the welding area is subjected to topological analysis based on the welding feature descriptor. This process aims to deeply analyze the spatial relationship and connection pattern between the welding points to obtain deeper welding quality information. Through topological analysis, it is possible to identify how the various welding points in the welding area are related to each other, and the impact of these associations on the stability and reliability of the entire welding structure. For example, in the manufacturing process of high-order computing power autonomous driving FPC (flexible printed circuit) sensor modules, the welding quality is directly related to whether the electrical connection between electronic components and circuit boards is firm and reliable, so it is crucial to accurately understand the topological structure of the welding area. After the graph transformation network completes the dynamic feature mapping and generates welding feature descriptors, these descriptors need to be used to perform more sophisticated topological analysis. This step involves using the welding feature descriptor as input data, calculating the relative position, distance, angle and other geometric parameters between different welding points through a specific algorithm, and establishing a mathematical model based on these parameters that can fully reflect the internal structure of the welding area, that is, the welding association map. The welding association diagram not only records the specific coordinates of each welding point, but more importantly, it depicts the connection path and strength distribution between each welding point, so that engineers can intuitively see which parts are key connection points and where there may be potential weak links. In practical applications, for example, for an autonomous driving FPC sensor module, assuming that a certain circuit requires particularly stable signal transmission, all welding points involved in the circuit can be found through topological structure analysis, and further check whether there are abnormal connections or weak areas in its welding association diagram that may affect signal integrity. If a low connection density or unexpected long-distance connection is detected around a welding point, this may be a problem in the welding process, such as poor welding caused by insufficient solder or overheating. At this point, corresponding improvement measures can be taken based on the detailed information provided by the welding association diagram to ensure that all welding points meet the best quality standards, thereby ensuring the performance and safety of the final product. In addition, in order to achieve efficient and accurate topological structure analysis, machine learning technology is usually combined to assist the analysis process. By training a classifier or clustering algorithm, the system can automatically learn the distinguishing features of normal and abnormal welding modes from a large number of welding feature descriptors, thereby improving the ability to understand complex welding structures. This not only speeds up the analysis, but also reduces the errors caused by human judgment. In short, by conducting a comprehensive and detailed topological structure analysis of the welding area and obtaining an accurate welding correlation diagram, a solid foundation is provided for subsequent quality control, and a direction is also pointed out for optimizing the welding process.

[0028] Step S5, inputting the welding association graph into a preset region growing algorithm for quality detection and grade classification to obtain a welding quality assessment report.

[0029] Specifically, after completing the topological structure analysis of the welding area and obtaining the welding association map, the next step is to input the welding association map into the preset region growing algorithm for quality inspection and grade classification. This process is to further evaluate the quality of welding and ensure that each welding point meets the established standards, thereby ensuring the safety and reliability of the entire welding structure. Specifically, the region growing algorithm is an image processing technology that gradually adds adjacent pixels with similar attributes (such as grayscale value, color, texture, etc.) to the same region starting from one or more seed points until all qualified pixels are included. In this application scenario, we are concerned about the connection strength and stability between welding points. Therefore, the algorithm will use the key welding points as seed points based on the information provided by the welding association map, and then determine which welding points should be classified into the same quality level based on the set rules and thresholds, such as the distance, angle and solder distribution between welding points. Taking the manufacturing of high-order computing power autonomous driving FPC sensor module as an example, assume that we have obtained a detailed welding association map of the welding area of ​​the module, which records the location of each welding point and its relationship with each other. Now, we input this map into the pre-configured region growing algorithm. The algorithm first selects some representative solder joints as initial seed points, which may be key connection locations previously identified through topological structure analysis. As the algorithm runs, it checks the solder joints around each seed point and determines whether these solder joints should be merged with the seed point into a larger region based on their physical characteristics (such as solder thickness, contact area, etc.) and electrical properties (such as resistance value). If a solder joint is found to not meet the expected quality standards, such as a cold solder joint or excessive solder, then this point will not be added to the current region, but will be marked for subsequent review. When all solder joints have been evaluated, the region growing algorithm will generate a detailed solder quality assessment report. This report not only contains the specific quality ratings of each solder joint, but may also point out those problem areas that need special attention. This is a very valuable reference for engineers because it can help them quickly locate potential quality risks and take corresponding corrective measures. For example, in the example of the autonomous driving FPC sensor module, if some solder joints are rated as low, this may mean that the signal transmission efficiency in the area will be affected, which in turn affects the performance of the entire system. At this point, engineers can adjust welding process parameters or replace problematic components based on the recommendations in the report to ensure that the final product meets strict quality requirements. In addition, by continuously applying this process, manufacturers can continuously optimize the welding process, improve production efficiency and improve product quality. In short, by combining the welding association map with the regional growing algorithm, effective monitoring and management of welding quality can be achieved, providing guarantees for the reliable operation of complex electronic equipment.

[0030] In a specific embodiment, the welding area of ​​the target FPC sensor module is scanned and imaged at multiple angles by a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area, including: Performing a full spectrum scan of the welding area through a preset multi-spectral imaging system to obtain an original spectral data sequence, wherein the original spectral data sequence includes welding information of a visible spectrum, a near infrared spectrum, and an ultraviolet spectrum; Performing band separation on the original spectral data sequence to obtain a multi-band spectral characteristic graph, and performing spectral enhancement processing on the multi-band spectral characteristic graph to obtain an enhanced spectral characteristic graph; Perform three-dimensional reconstruction based on the enhanced spectral characteristic graph to obtain a welding stereoscopic spectrum graph; Based on the spectral registration network, the welding stereo spectrum image is spatially registered to obtain a registered spectral data set, and the registered spectral data set is feature fused by a spectral feature fusion algorithm to obtain a fused spectral feature image; Performing end-member analysis on the fused spectral characteristic graph by using a spectral unmixing algorithm to obtain a welding material component distribution graph; and performing geometric correction on the welding material component distribution graph to obtain a corrected component graph; wherein the welding material component distribution graph includes solder composition, substrate characteristics and interface structure information; The corrected component map is subjected to dimensionality reduction and denoising processing to obtain a multidimensional spectral image set of the welding area.

[0031] Specifically, in the process of multi-angle scanning and imaging of the welding area of ​​the target FPC sensor module, a series of complex steps are implemented through the preset multispectral imaging system to ensure that the most detailed and accurate information of the welding area can be captured. This process begins with full-spectrum scanning, that is, using a multispectral imaging system to fully cover the welding area and obtain a sequence of raw spectral data from the visible spectrum, near-infrared spectrum to the ultraviolet spectrum. These raw spectral data not only contain the color information of the welding surface, but also record the ability of the material to reflect or transmit light of different wavelengths, providing a rich basis for subsequent analysis. Specifically, in the application scenario of the FPC sensor module for autonomous driving, when engineers need to evaluate the welding quality of a new batch, they will start the multispectral imaging system, which is equipped with a variety of filters and sensors to complete the full spectrum scanning of the welding point in a short time. Each scan generates a large amount of raw spectral data, which is organized into an ordered data sequence, and each data point represents the reflectivity or transmittance of a part of the welding area at a specific wavelength. For example, for the interface between solder and substrate, different materials may show similar color characteristics in the visible spectrum range, but show obvious differences in the near-infrared or ultraviolet spectrum, which helps to identify potential problems such as cold solder joints or uneven solder distribution. To further process these raw spectral data sequences, the next step is to perform band separation on the raw spectral data sequences, with the goal of decomposing the entire spectral range into multiple independent bands to obtain a multi-band spectral feature map. In this process, the image of each band can highlight the welding characteristics within a specific wavelength range. Then, in order to enhance the key features in the image, these multi-band spectral feature maps are spectrally enhanced to make the subtle differences in the welding area more obvious by adjusting the contrast, brightness or other parameters. For example, after spectral enhancement, tiny cracks or pores that were originally difficult to detect may become clearly visible, which is crucial to ensure the quality of welding. Based on the enhanced spectral feature map, the next step is the 3D reconstruction stage, where the goal is to construct a three-dimensional spectral map of the welding. This step involves using computer vision technology to infer the spatial structure of the welding area based on the two-dimensional spectral feature maps under various viewing angles. 3D reconstruction not only provides a plan view of the weld, but also shows its height variations, helping engineers to more intuitively understand the morphological characteristics of the weld. For example, in some cases, the height difference of the weld may be caused by spatter generated during the welding process, and 3D reconstruction can accurately reveal this microstructure so that appropriate corrective measures can be taken. Once the weld stereo spectrogram is obtained, it needs to be spatially registered to ensure that data from different viewing angles or time points can be accurately aligned. This is achieved through a spectral registration network, which can automatically identify and correct deviations caused by changes in imaging conditions (such as camera position, angle, lighting, etc.).The registered spectral data set lays the foundation for the next step of feature fusion. On this basis, the information from different bands is integrated through the spectral feature fusion algorithm to generate a comprehensive fused spectral feature map. This step enhances the overall information content of the image, while reducing redundant data, making the welding features more concentrated and significant. In order to deeply understand the composition and distribution of the welding material, the spectral unmixing algorithm is used to perform end member analysis on the fused spectral feature map to obtain the welding material component distribution map. This process is similar to spectral analysis in chemical analysis. By identifying different spectral signatures, the main components in the welding material and their spatial distribution can be determined. For example, the solder composition, substrate characteristics and interface structure information can be clearly displayed in this picture. However, since some geometric distortions are inevitably introduced in the actual imaging process, it is also necessary to perform geometric correction on the welding material component distribution map to ensure the authenticity and accuracy of the final result. The corrected component map can provide engineers with first-hand information on welding quality and help them quickly locate the problem. Finally, in order to facilitate subsequent quality inspection and analysis, the corrected component map needs to be reduced in dimension and denoised to obtain a multidimensional spectral image set. For example, in the manufacturing process of FPC sensor modules for autonomous driving, multidimensional spectral image sets after dimensionality reduction and denoising can be used in automated inspection systems to quickly screen out qualified and unqualified products, greatly improving production efficiency and product quality control. In summary, a series of carefully designed operations such as full-spectrum scanning, band separation, spectral enhancement, three-dimensional reconstruction, spatial registration, feature fusion, end-member analysis, geometric correction, and dimensionality reduction and denoising of the welding area have been implemented to achieve a comprehensive evaluation of the welding quality of the FPC sensor module. This method not only improves the detection accuracy, but also provides strong support for optimizing the welding process and improving product reliability, ensuring the stable operation of the autonomous driving system.

[0032] In a specific embodiment, the method of extracting features from the multidimensional spectral image set by a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector includes: The spatial spectrum feature extraction is performed on the multi-dimensional spectral image set by a preset convolutional neural network to obtain a spatial spectrum feature map; wherein the spatial spectrum feature map includes spatial texture information and spectral component information of the welding area; Performing multi-scale feature fusion on the spatial spectrum feature map through a pyramid pooling module to obtain a multi-scale feature representation, and performing local binary pattern encoding on the multi-scale feature representation to obtain an LBP feature map; Performing global context modeling on the LBP feature map based on a Transformer network to obtain a global context feature vector; Performing feature compression on the global context feature vector by using a local sensitive hash coding technique to obtain a sparse feature vector, and using the sparse feature vector as a welding feature vector of the welding area; Relationship modeling is performed based on the sparse feature vector to obtain a welding feature relationship diagram, and a welding feature map is constructed based on the welding feature relationship diagram; wherein the welding feature map includes welding intensity distribution data, welding depth distribution data and welding surface flatness data.

[0033] Specifically, in order to achieve in-depth analysis of the welding area of ​​the FPC sensor module, the method mentioned in the claims introduces a preset convolutional neural network (CNN) to extract features from a multidimensional spectral image set. This process is accomplished by inputting the multidimensional spectral image set into a pre-trained convolutional neural network that can recognize and learn complex patterns in the image to obtain a spatial spectral feature map containing spatial texture information and spectral component information of the welding area. In this process, the multi-layer structure of the convolutional neural network enables it to gradually abstract high-level features (such as shape, texture) from low-level features (such as edges and corners), which is crucial for understanding the quality of welding. For example, in the manufacturing process of autonomous driving sensors, engineers may use this method to ensure that each welding point achieves optimal performance. When a multidimensional spectral image set is fed into the convolutional neural network, the network automatically processes the images and extracts the features that are most critical to the evaluation of welding quality. For example, if a batch of welding points looks normal under visible light, but shows abnormal reflectivity changes in the spectral image at a specific wavelength, this may mean that the solder is not fully fused or there are other quality problems. Convolutional neural networks have the ability to capture such subtle changes and encode them as part of the spatial-spectral feature map. Next, in order to further enhance the feature representation, the pyramid pooling module is used in the method to perform multi-scale feature fusion on the spatial-spectral feature map. This is because welding defects may appear at different scales, from macroscopic shape irregularities to microscopic material distribution differences, so a mechanism is needed to integrate these cross-scale information. The pyramid pooling module obtains a richer and more comprehensive multi-scale feature representation by sampling feature maps at multiple scales and combining the results. Then, local binary pattern (LBP) encoding is applied to these multi-scale feature representations to generate LBP feature maps. LBP is an effective method for describing the local texture of an image. It can highlight the details of the weld surface and reveal potential problems even at very small scales. Based on the obtained LBP feature map, the Transformer network is then used to model the global context. Transformer was originally designed as a model architecture for natural language processing, but it is also applicable to computer vision tasks, especially when the relationship between different parts of the image needs to be considered. In the application scenario of welding quality inspection, Transformer can be used to capture global context information within the welding area, that is, the interaction between the welding point and its surrounding environment. For example, whether a welding point is properly bonded to the surrounding substrate, or whether there is a risk of cracks due to stress concentration, these issues can be more accurately evaluated through global context modeling. Ultimately, this step produces a global context feature vector that condenses comprehensive information about the welding area.However, such a feature vector may be very large and contain too much data, which is not conducive to subsequent rapid processing or storage. Therefore, the method also proposes to use local sensitive hashing (LSH) coding technology to compress the global context feature vector to generate a sparse feature vector. LSH coding is a technology that can reduce the data dimension while maintaining similarity. It allows even high-dimensional features to be effectively compressed into sparse representations while retaining the main characteristics of the original features. The sparse feature vector obtained in this way is not only easy to process, but also can still accurately represent the key attributes of the welding area. It can be used as a welding feature vector for the next step of analysis. Finally, relationship modeling is performed based on the sparse feature vector to construct a welding feature map. This map is not just a static image, but a dynamic relationship network, in which each node represents a welding feature and the edge represents the relationship between them. For example, welding intensity distribution data, welding depth distribution data, and welding surface flatness data are all integrated into this map to form a complete description of welding quality. In this way, engineers can gain a comprehensive view of the health of solder joints and take appropriate measures to improve production processes or resolve issues found, ensuring that each FPC sensor module meets stringent performance standards and ensures the safety and reliability of autonomous driving systems.

[0034] In a specific embodiment, the method of dynamically mapping the welding feature map through a map transformation network to obtain a welding feature descriptor includes: Performing multi-resolution decomposition on the welding feature atlas through a preset atlas transformation network to obtain a welding feature sub-atlas set; Based on the bidirectional long short-term memory network, the welding feature sub-atlas set is subjected to time series feature analysis to obtain a time series feature sequence, and the time series feature sequence is subjected to structured analysis through the graph attention network to obtain a welding dynamic feature graph; Performing Fourier spectrum analysis on the welding dynamic characteristic graph to obtain a frequency domain characteristic spectrum; Extract key features from the frequency domain feature spectrum using an adaptive feature selection algorithm to obtain a welding feature description vector; Performing topological encoding on the welding feature description vector based on a graph isomorphic network to obtain a structured feature encoding; The structured feature code is transformed and mapped by a graph mapping algorithm to obtain a welding feature descriptor; wherein the welding feature descriptor includes an interface bonding state, a void distribution density and a metal interconnection strength.

[0035] Specifically, in order to deeply analyze the welding quality, the claims propose a method for dynamically mapping the welding feature map through a map transformation network to obtain a more detailed and accurate welding feature descriptor. The process begins by inputting the welding feature map into a preset map transformation network, which first performs multi-resolution decomposition to obtain a series of welding feature sub-map sets. These sub-map sets not only represent welding features at different scales, but also provide a basis for subsequent temporal feature analysis. For example, in the manufacturing process of FPC (flexible printed circuit) sensor modules, engineers may use this method to ensure that each welding point meets the optimal quality standards. When a welding feature map is fed into the map transformation network, the network decomposes it into multiple sub-maps according to different resolution levels. Each sub-map focuses on features at a specific scale, from the macroscopic overall morphology to the microscopic surface texture, which ensures that all levels of welding features are fully considered. This step is crucial for identifying subtle defects that may be overlooked at a single scale, such as tiny voids or poor interface bonding. Next, in order to understand the changes of welding features over time, the method introduces a bidirectional long short-term memory network (BiLSTM) to perform temporal feature analysis on the welding feature sub-graph set. BiLSTM is a neural network architecture that is particularly suitable for processing sequence data. It can capture the evolution pattern of welding features in the manufacturing process and generate a temporal feature sequence. At the same time, in order to further reveal the intrinsic connection between these temporal features, the graph attention network (GAT) is used to perform a structured analysis of the temporal feature sequence to obtain a welding dynamic feature graph. The graph attention network allows the model to focus on the most relevant feature nodes and their connections, thereby better simulating the complex relationships within the welding area. For example, during the welding process, the melting, cooling, and solidification stages of the solder will affect the final welding quality, and GAT can help us understand and predict the impact of these dynamic changes more accurately. Based on the welding dynamic feature graph obtained above, Fourier spectrum analysis is performed on it. Fourier transform is a technique for converting signals from the time domain to the frequency domain. In this way, the frequency domain feature spectrum can be extracted from the welding dynamic feature graph. This transformation allows us to examine the welding characteristics from a frequency perspective, which helps to discover periodic or quasi-periodic behaviors that are not easily noticeable in the time domain. For example, if there is some regular vibration or temperature fluctuation during the welding process, these phenomena will appear as peaks at specific frequencies in the frequency domain feature spectrum. Such information is very valuable for diagnosing welding problems because it can help engineers identify potential sources of problems and take appropriate preventive measures. Subsequently, in order to refine the obtained frequency domain feature spectrum, an adaptive feature selection algorithm is used to extract key features from the frequency domain feature spectrum to obtain a welding feature description vector. The adaptive feature selection algorithm aims to find the key features that best characterize the welding quality and remove redundant or irrelevant information.This means that the final welding feature description vector will only contain those attributes that are critical to evaluating welding quality, such as interface bonding state, void distribution density, and metal interconnection strength. Doing so can not only improve the efficiency of subsequent analysis, but also enhance the reliability of the results. Finally, in order to fully represent the information in the welding feature description vector, the method proposes to use a graph isomorphism network (GIN) to perform topological structure encoding on the welding feature description vector to obtain a structured feature encoding. A graph isomorphism network is a deep learning model for learning graph structured data. It can encode node features while maintaining the topological properties of the graph. In this way, even in the face of complex welding feature maps, its internal structural characteristics can be accurately captured. Then, the structured feature encoding is transformed and mapped by a graph mapping algorithm, and finally a welding feature descriptor is obtained. The welding feature descriptor is a highly condensed representation that comprehensively reflects various key attributes of the welding area, including but not limited to the interface bonding state, void distribution density, and metal interconnection strength. This not only provides a solid foundation for the quantitative evaluation of welding quality, but also provides valuable guidance for optimizing welding process parameters, ensuring the reliability and performance of FPC sensor modules, especially in application scenarios such as autonomous driving sensors that have extremely high safety requirements.

[0036] In a specific embodiment, the topological structure analysis of the welding area based on the welding feature descriptor to obtain a welding association graph includes: The welding feature descriptor is locally deconstructed by a preset welding distribution analysis algorithm to obtain a local feature distribution map of the welding area, and the local feature distribution map is subjected to feature connection analysis to obtain welding feature local connection data; wherein the welding feature local connection data includes position information, connection weight information and node type information of welding nodes; Based on the welding feature local connection data, a global network model is performed on the welding area to obtain a global topological structure diagram of the welding area, and feature mapping is performed on the global topological structure diagram to obtain an embedded topological feature vector; Inferring the dynamic adjacency relationship of the embedded topological feature vector through a topological optimization network to obtain dynamic feature association data between welding areas, and performing feature aggregation on the dynamic feature association data to obtain a welding cluster feature distribution map, which includes spatial association information of welding group structure and defect characteristics; Performing geometric structural decomposition on the welding cluster feature distribution map by using a geometric association analysis algorithm to obtain a geometric association map of the welding area, and performing high-order feature mapping on the geometric association map to obtain a welding geometric feature description map; Based on the welding geometric feature description diagram, multi-layer relationship modeling is performed on the dynamic characteristics of the welding area to obtain a welding dynamic feature diagram; The welding dynamic feature map and the geometric association map are jointly encoded by a high-order feature fusion algorithm to obtain a welding association map.

[0037] Specifically, in order to conduct a comprehensive and detailed topological structure analysis of the welding area, a method based on welding feature descriptors is proposed in the claims, aiming to obtain a welding association map that can accurately reflect the welding quality through multi-step processing and modeling. This process begins with the preset welding distribution analysis algorithm to perform local structural deconstruction on the welding feature descriptor to obtain the local feature distribution map of the welding area, and further perform feature connection analysis on these local feature distribution maps to obtain welding feature local connection data. In the application scenario of FPC (flexible printed circuit) sensor modules, engineers can use this method to ensure that each welding point meets the best quality standards. Specifically, when the welding feature descriptor is fed into the welding distribution analysis algorithm, the algorithm will decompose the complex welding features into multiple local structures and generate a local feature distribution map. Each local feature distribution map records the location information, connection weight information and node type information of the welding node in detail, which allows engineers to clearly see the specific situation of each welding point and its surrounding environment. For example, in the manufacturing process of autonomous driving sensors, if the solder around a welding point is unevenly distributed or there is a cold solder joint, these problems will appear in the local feature distribution map. Next, by performing feature connection analysis on these local feature distribution maps, the connection relationship between different welding nodes can be determined, that is, the local connection data of welding features. This step helps to reveal the physical connection strength and stability between welding points. Based on the obtained local connection data of welding features, the next step is to perform global network modeling on the welding area to construct a global topological structure map of the welding area. This global topological structure map not only shows the direct connection between welding points, but also reflects the spatial layout and overall structure of the entire welding area. In order to better understand and analyze these complex relationships, the global topological structure map will be feature mapped to generate an embedded topological feature vector. The embedded topological feature vector is a compact and expressive data representation that condenses the key information about the topological structure of the welding area, making the subsequent dynamic adjacency relationship inference more efficient and accurate. Then, the embedded topological feature vector is dynamically inferred through the topological optimization network to obtain the dynamic feature association data between welding areas. The topological optimization network is a deep learning model specially designed to learn and predict the relationship between nodes in graph structure data. At this stage, the network will infer the dynamic relationship between them based on the connection pattern and feature similarity between welding points, such as whether there is stress concentration or heat conduction path. Subsequently, these dynamic feature association data are aggregated to generate a welding cluster feature distribution map, which includes the spatial association information of the welding group structure and defect characteristics.For example, some welding points may form specific clustering patterns due to the influence of material properties or process parameters. These patterns may be related to welding quality, and the welding cluster feature distribution map can help identify these patterns so that targeted improvement measures can be taken. In order to further understand the geometric structure of the welding area, the geometric association analysis algorithm is used to decompose the welding cluster feature distribution map to obtain the geometric association map of the welding area. The geometric association map provides detailed information about the spatial position, shape and relative arrangement of the welding points, which is crucial for evaluating the flatness of the welding surface and the distribution of solder. Then, the welding geometric feature description map is generated by performing high-order feature mapping on the geometric association map. This map not only contains the geometric properties of the welding point itself, but also integrates its interaction with the surrounding environment, providing a basis for deeper analysis. Finally, based on the welding geometric feature description map, multi-layer relationship modeling is performed on the dynamic characteristics of the welding area to obtain the welding dynamic feature map. This step involves the complex interaction between various factors inside the welding area, such as temperature changes, mechanical stress, and material aging. These dynamic characteristics have an important impact on the reliability in long-term use. The welding dynamic feature map and geometric association map are jointly encoded by a high-order feature fusion algorithm to finally obtain a welding association map. The welding association map is a comprehensive representation that integrates the geometric structure and dynamic behavior of the welding area. It can provide engineers with a comprehensive and detailed basis for welding quality assessment, helping them to quickly locate potential problems and optimize production processes, ensuring that each FPC sensor module can meet strict performance standards, especially in application environments such as autonomous driving systems that have extremely high safety requirements.

[0038] In a specific embodiment, the welding association graph is input into a preset region growing algorithm for quality detection and grade classification to obtain a welding quality assessment report, including: The seed points of the welding association graph are located by a regional growth seed selection algorithm to obtain a welding region seed point set, and the regional growth factor is calculated for the welding region seed point set to obtain regional growth feature data; wherein the regional growth feature data includes welding connection strength, welding boundary gradient and welding region morphological characteristics; Based on the regional growth feature data, the welding region is adaptively expanded to obtain a welding region growth map, and the welding region growth map is extracted from the boundary features by a regional boundary tracking algorithm to obtain a welding region boundary feature set; Performing quality inspection and grading on the welding area boundary feature set according to a preset quality assessment criterion to obtain a welding quality grade distribution map; Based on the deep feature analysis network, the welding quality grade distribution map is subjected to deep feature analysis to obtain a welding quality feature vector, and the welding quality feature vector is subjected to hierarchical evaluation through a multi-level evaluation algorithm to obtain welding quality evaluation data; wherein the welding quality evaluation data includes a welding strength score, a welding depth score and a welding surface score; The welding quality assessment data is structured and content optimized through an assessment report generation algorithm to obtain a welding quality assessment report; wherein the welding quality assessment report includes quality grade distribution statistics, defect type analysis and process parameter recommendations.

[0039] Specifically, in order to achieve quality inspection and grading of welding areas, a method is proposed in the claims, which inputs the welding association map into a preset regional growth algorithm. This process first requires the positioning of the seed points of the welding association map through a technique called regional growth seed selection algorithm. In application scenarios such as flexible printed circuit (FPC) sensor module manufacturing, engineers can use this method to ensure that each welding point meets the best quality standards. Specifically, the regional growth seed selection algorithm automatically identifies and selects key positions that can represent different welding characteristics as seed points, which constitute a set of welding area seed points. Once the location of the seed point is determined, the next step is to calculate the regional growth factor of these welding area seed points to obtain regional growth feature data, including information such as welding connection strength, welding boundary gradient and welding area morphological characteristics. For example, at the key welding points in the FPC module, if they show abnormal connection strength or irregular morphological characteristics, then this information will be accurately recorded. After obtaining detailed regional growth feature data, it becomes possible to perform adaptive regional expansion of the welding area based on these data. Adaptive region expansion refers to dynamically determining the direction and range of growth based on the features around each seed point, and finally forming a welding region growth map. In this process, in order to more accurately understand the edge of the welding area, the regional boundary tracking algorithm is also used to extract the boundary features of the welding region growth map, and then obtain the welding region boundary feature set. This feature set contains information about the shape, continuity and any possible defects of the welding boundary, which is crucial for subsequent quality assessment. For example, in the manufacturing process of autonomous driving sensors, if there are cracks or discontinuities at the boundary of a certain welding point, these problems will be clearly displayed in the welding region boundary feature set, allowing engineers to discover and take corrective measures in time. Subsequently, the welding region boundary feature set is analyzed according to the preset quality assessment criteria to perform quality inspection and grade classification, and generate a welding quality grade distribution map. This step involves a series of strict standards and thresholds to determine whether the welding quality meets expectations. The welding quality grade distribution map not only shows the quality differences between different welding points, but also provides a visual reference for subsequent improvements. For example, in an FPC module, some welding points may show a low quality level due to improper process parameter settings. The welding quality grade distribution map can intuitively show which areas need special attention. In order to further understand the nature of welding quality, the deep feature analysis network is used to perform deep feature analysis on the welding quality grade distribution map to obtain the welding quality feature vector. The deep feature analysis network is a complex machine learning model that can mine hidden patterns and laws from large amounts of data.The welding quality feature vector is the result of this analysis, which comprehensively reflects various aspects of welding quality, such as welding strength score, welding depth score and welding surface score. By applying a multi-level evaluation algorithm to the welding quality feature vector, a comprehensive hierarchical evaluation of welding quality can be performed to obtain welding quality evaluation data. This evaluation data is a summary of the entire welding process. It quantifies different aspects of welding quality and helps engineers make more scientific and reasonable decisions. Finally, in order to make these evaluation results more practical and readable, the welding quality evaluation data is structured and optimized through the evaluation report generation algorithm to form a welding quality evaluation report. This report is not just a collection of numbers and charts, but more importantly, it contains quality grade distribution statistics, defect type analysis, and process parameter recommendations for the problems found. For example, on the production line of FPC sensor modules, if the welding quality evaluation report shows that a batch of products generally has insufficient welding strength, then the process parameter recommendations given in the report can guide engineers to adjust the welding temperature, time or other related factors, thereby improving product quality and ensuring that every FPC sensor module shipped can meet strict performance requirements, especially in application environments such as autonomous driving systems that have extremely high safety requirements.

[0040] In a specific embodiment, the adaptive region expansion of the welding region based on the region growth feature data to obtain a welding region growth map includes: Performing dynamic threshold calculation on the regional growth feature data by using a preset regional growth threshold adaptive algorithm to obtain a regional growth threshold set, and locally optimizing the regional growth threshold based on the morphological characteristics of the welding area to obtain an optimized growth threshold set; Based on the optimized growth threshold set, the welding area is expanded in multiple directions to obtain a regional expansion direction map, and the expansion direction map is connected to mark the domain through a regional connectivity analysis algorithm to obtain a welding area connection mark map; The growth constraint calculation is performed on the welding area connected labeling graph by using a preset regional growth constraint algorithm to obtain regional growth constraint data. Based on the regional competition mechanism, multi-region competition growth is performed on the regional growth constraint data to obtain a regional competition growth map, and similar regions of the regional competition growth map are merged through a regional merging algorithm to obtain a welding region growth map.

[0041] Specifically, in order to achieve adaptive regional expansion of the welding area, a method based on regional growth feature data is proposed in the claims, which aims to generate a welding region growth map through a series of carefully designed steps. This process first involves a preset regional growth threshold adaptive algorithm to perform dynamic threshold calculation on the regional growth feature data to obtain a regional growth threshold set. The key to this stage is to dynamically adjust the growth threshold around each seed point based on information such as welding connection strength, welding boundary gradient and welding area morphological characteristics to ensure that they can accurately reflect the changes in local welding quality. For example, in the manufacturing process of FPC (flexible printed circuit) sensor modules, engineers may encounter significant quality differences between different welding points. For those areas with high welding connection strength and regular morphological characteristics, relatively loose growth thresholds can be set; while for those areas that may have defects or irregular morphology, more stringent thresholds are required to avoid misjudgment. Therefore, through the regional growth threshold adaptive algorithm, the threshold can be automatically adjusted according to the specific welding characteristics, making the growth process more accurate and reliable. After determining the initial regional growth threshold, it is also necessary to locally optimize these thresholds based on the welding area morphological characteristics to obtain an optimized growth threshold set. This step is to further improve the pertinence of the growth threshold and ensure that the growth conditions around each welding point can be optimally configured. Next, the welding area is multi-directionally expanded based on the optimized growth threshold set to generate a regional expansion direction map. Multi-directional expansion calculation refers to starting from each seed point and gradually expanding along multiple possible directions until the set growth threshold is reached. In this process, the direction and distance of each expansion are recorded to form a detailed regional expansion direction map. In order to ensure the connectivity between each expansion direction, the regional connectivity analysis algorithm is used to process the expansion direction map to obtain the welding area connectivity mark map. The regional connectivity analysis algorithm can identify and mark all interconnected regions, so that the final generated welding area growth map not only covers all key welding points, but also maintains the consistency and integrity of the overall structure. For example, in the manufacture of autonomous driving sensors, if there are multiple connected welding points around a welding point, these welding points will be correctly classified as the same connected domain for subsequent analysis. In order to control the expansion range of the welding area and prevent overgrowth or missing important areas, the welding area connectivity mark map is then calculated by the preset regional growth constraint algorithm to obtain regional growth constraint data. The region growth constraint algorithm introduces a series of rules and constraints, such as maximum growth radius, minimum connected area, etc., to ensure that the expansion of each welding region is within a reasonable range. These constraints help maintain the overall balance of the welding region and avoid affecting the normal evaluation of other parts due to excessive expansion of some local areas.For example, when the solder distribution around a welding point is particularly dense, if it is not constrained, it may cause the area to expand abnormally, covering up other potential problems. The regional growth constraint algorithm can effectively avoid this situation and make the evaluation of the entire welding area more scientific and reasonable. Finally, in order to simulate the actual welding situation, a method based on the regional competition mechanism is used to perform multi-region competition growth on the regional growth constraint data to obtain a regional competition growth map. The regional competition mechanism imitates the competitive relationship between different biological groups in nature, that is, adjacent welding areas will compete during the expansion process, and only those areas that are more in line with the growth conditions can continue to expand. This mechanism can better reflect the complex interactions in the actual welding process and improve the authenticity of the evaluation results. Subsequently, the welding area growth map is obtained by merging similar regions of the regional competition growth map. Similar region merging is to merge those connected domains with similar characteristics into larger regions, which reduces redundant information and simplifies subsequent analysis. For example, in the FPC sensor module, if the connection between several welding points is very tight and the characteristics are similar, they will be merged into a larger welding area, which not only improves efficiency but also helps to more intuitively display the overall quality of welding. In summary, through a series of steps such as dynamic threshold calculation, multi-directional expansion calculation, connectivity analysis, growth constraint calculation, and regional competitive growth of regional growth feature data, the adaptive expansion of the welding area is realized, and finally the welding area growth map is obtained. This method not only improves the accuracy of welding quality assessment, but also provides engineers with a comprehensive and detailed tool to help them quickly locate problems and optimize production processes, ensuring that every FPC sensor module shipped meets strict quality standards, especially in application environments such as autonomous driving systems that have extremely high safety requirements.

[0042] The above describes the welding quality detection method of the high-order computing power autonomous driving FPC sensor module in the embodiment of the present invention. The following describes the welding quality detection system of the high-order computing power autonomous driving FPC sensor module in the embodiment of the present invention. Please refer to Figure 2 , an embodiment of the welding quality detection system of the high-order computing power autonomous driving FPC sensor module in the embodiment of the present invention includes: An imaging module 21 is used to perform multi-angle scanning imaging on the welding area of ​​the target FPC sensor module through a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area; An extraction module 22 is used to extract features from the multidimensional spectral image set through a preset convolutional neural network to obtain a welding feature vector of the welding area, and construct a welding feature map based on the welding feature vector; A mapping module 23, used for performing dynamic feature mapping on the welding feature map through a map transformation network to obtain a welding feature descriptor; An analysis module 24, configured to perform a topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association diagram; The classification module 25 is used to input the welding association diagram into a preset region growing algorithm to perform quality grade classification to obtain a welding quality assessment report.

[0043] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0044] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0046] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0047] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0048] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0049] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A welding quality detection method for a high-order computing power autonomous driving FPC sensor module, characterized in that: The following steps are involved: The welding area of ​​the target FPC sensor module is scanned and imaged at multiple angles by a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area; Extracting features from the multidimensional spectral image set by a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector; Performing dynamic feature mapping on the welding feature map through a map transformation network to obtain a welding feature descriptor; Performing a topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association graph; The welding association diagram is input into a preset region growing algorithm for quality inspection and grade classification to obtain a welding quality assessment report.

2. The welding quality detection method of the high-order computing power autonomous driving FPC sensor module according to claim 1 is characterized in that: The multi-dimensional spectral image set of the welding area of ​​the target FPC sensor module is obtained by scanning and imaging the welding area of ​​the target FPC sensor module at multiple angles through a preset multi-spectral imaging system, including: Performing a full spectrum scan of the welding area through a preset multi-spectral imaging system to obtain an original spectral data sequence, wherein the original spectral data sequence includes welding information of a visible spectrum, a near infrared spectrum, and an ultraviolet spectrum; Performing band separation on the original spectral data sequence to obtain a multi-band spectral characteristic graph, and performing spectral enhancement processing on the multi-band spectral characteristic graph to obtain an enhanced spectral characteristic graph; Perform three-dimensional reconstruction based on the enhanced spectral characteristic graph to obtain a welding stereoscopic spectrum graph; Based on the spectral registration network, the welding stereo spectrum image is spatially registered to obtain a registered spectral data set, and the registered spectral data set is feature fused by a spectral feature fusion algorithm to obtain a fused spectral feature image; Performing end-member analysis on the fused spectral characteristic graph by a spectral unmixing algorithm to obtain a welding material component distribution graph, and performing geometric correction on the welding material component distribution graph to obtain a corrected component graph; wherein the welding material component distribution graph includes solder composition, substrate characteristics and interface structure information; The corrected component map is subjected to dimensionality reduction and denoising processing to obtain a multidimensional spectral image set of the welding area.

3. The welding quality detection method of the high-order computing power autonomous driving FPC sensor module according to claim 1 is characterized in that: The method of extracting features from the multidimensional spectral image set by a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector, includes: The spatial spectrum feature extraction is performed on the multi-dimensional spectral image set by a preset convolutional neural network to obtain a spatial spectrum feature map; wherein the spatial spectrum feature map includes spatial texture information and spectral component information of the welding area; Performing multi-scale feature fusion on the spatial spectrum feature map through a pyramid pooling module to obtain a multi-scale feature representation, and performing local binary pattern encoding on the multi-scale feature representation to obtain an LBP feature map; Performing global context modeling on the LBP feature map based on a Transformer network to obtain a global context feature vector; Performing feature compression on the global context feature vector by using a local sensitive hash coding technique to obtain a sparse feature vector, and using the sparse feature vector as a welding feature vector of the welding area; Relationship modeling is performed based on the sparse feature vector to obtain a welding feature relationship diagram, and a welding feature map is constructed based on the welding feature relationship diagram; wherein the welding feature map includes welding intensity distribution data, welding depth distribution data and welding surface flatness data.

4. The welding quality detection method of the high-order computing power autonomous driving FPC sensor module according to claim 1 is characterized in that: The method of dynamically mapping the welding feature map through a map transformation network to obtain a welding feature descriptor includes: Performing multi-resolution decomposition on the welding feature atlas through a preset atlas transformation network to obtain a welding feature sub-atlas set; Based on the bidirectional long short-term memory network, the welding feature sub-atlas set is subjected to time series feature analysis to obtain a time series feature sequence, and the time series feature sequence is subjected to structured analysis through the graph attention network to obtain a welding dynamic feature graph; Performing Fourier spectrum analysis on the welding dynamic characteristic graph to obtain a frequency domain characteristic spectrum; Extract key features from the frequency domain feature spectrum using an adaptive feature selection algorithm to obtain a welding feature description vector; Performing topological encoding on the welding feature description vector based on a graph isomorphic network to obtain a structured feature encoding; The structured feature code is transformed and mapped by a graph mapping algorithm to obtain a welding feature descriptor; wherein the welding feature descriptor includes an interface bonding state, a void distribution density and a metal interconnection strength.

5. The welding quality detection method of the high-order computing power autonomous driving FPC sensor module according to claim 1 is characterized in that: The performing topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association graph includes: The welding feature descriptor is locally deconstructed by a preset welding distribution analysis algorithm to obtain a local feature distribution map of the welding area, and the local feature distribution map is subjected to feature connection analysis to obtain welding feature local connection data; wherein the welding feature local connection data includes position information, connection weight information and node type information of welding nodes; Based on the welding feature local connection data, a global network model is performed on the welding area to obtain a global topological structure diagram of the welding area, and feature mapping is performed on the global topological structure diagram to obtain an embedded topological feature vector; Inferring the dynamic adjacency relationship of the embedded topological feature vector through a topological optimization network to obtain dynamic feature association data between welding areas, and performing feature aggregation on the dynamic feature association data to obtain a welding cluster feature distribution map, which includes spatial association information of welding group structure and defect characteristics; Performing geometric structural decomposition on the welding cluster feature distribution map by using a geometric association analysis algorithm to obtain a geometric association map of the welding area, and performing high-order feature mapping on the geometric association map to obtain a welding geometric feature description map; Based on the welding geometric feature description diagram, multi-layer relationship modeling is performed on the dynamic characteristics of the welding area to obtain a welding dynamic feature diagram; The welding dynamic feature map and the geometric association map are jointly encoded by a high-order feature fusion algorithm to obtain a welding association map.

6. The welding quality detection method of the high-order computing power autonomous driving FPC sensor module according to claim 1 is characterized in that: The welding association graph is input into a preset region growing algorithm for quality detection and grade classification to obtain a welding quality assessment report, including: The seed points of the welding association graph are located by a regional growth seed selection algorithm to obtain a welding region seed point set, and the regional growth factor is calculated for the welding region seed point set to obtain regional growth feature data; wherein the regional growth feature data includes welding connection strength, welding boundary gradient and welding region morphological characteristics; Based on the regional growth feature data, the welding region is adaptively expanded to obtain a welding region growth map, and the welding region growth map is extracted from the boundary features by a regional boundary tracking algorithm to obtain a welding region boundary feature set; Performing quality inspection and grading on the welding area boundary feature set according to a preset quality assessment criterion to obtain a welding quality grade distribution map; Based on the deep feature analysis network, the welding quality grade distribution map is subjected to deep feature analysis to obtain a welding quality feature vector, and the welding quality feature vector is subjected to hierarchical evaluation through a multi-level evaluation algorithm to obtain welding quality evaluation data; wherein the welding quality evaluation data includes a welding strength score, a welding depth score and a welding surface score; The welding quality assessment data is structured and content optimized through an assessment report generation algorithm to obtain a welding quality assessment report; wherein the welding quality assessment report includes quality grade distribution statistics, defect type analysis and process parameter recommendations.

7. The welding quality detection method of the high-order computing power autonomous driving FPC sensor module according to claim 6 is characterized in that: The step of adaptively expanding the welding region based on the regional growth feature data to obtain a welding region growth map comprises: Performing dynamic threshold calculation on the regional growth feature data by using a preset regional growth threshold adaptive algorithm to obtain a regional growth threshold set, and locally optimizing the regional growth threshold based on the morphological characteristics of the welding area to obtain an optimized growth threshold set; Based on the optimized growth threshold set, the welding area is expanded in multiple directions to obtain a regional expansion direction map, and the expansion direction map is connected to mark the domain through a regional connectivity analysis algorithm to obtain a welding area connection mark map; The growth constraint calculation is performed on the welding area connected labeling graph by using a preset regional growth constraint algorithm to obtain regional growth constraint data. Based on the regional competition mechanism, multi-region competition growth is performed on the regional growth constraint data to obtain a regional competition growth map, and similar regions of the regional competition growth map are merged through a regional merging algorithm to obtain a welding region growth map.

8. A welding quality detection system for a high-order computing power autonomous driving FPC sensor module, characterized in that: include: An imaging module is used to perform multi-angle scanning imaging on the welding area of ​​the target FPC sensor module through a preset multi-spectral imaging system to obtain a multi-dimensional spectral image set of the welding area; An extraction module, used for performing feature extraction on the multidimensional spectral image set through a preset convolutional neural network to obtain a welding feature vector of the welding area, and constructing a welding feature map based on the welding feature vector; A mapping module, used for performing dynamic feature mapping on the welding feature map through a map transformation network to obtain a welding feature descriptor; An analysis module, configured to perform a topological structure analysis on the welding area based on the welding feature descriptor to obtain a welding association diagram; The classification module is used to input the welding association diagram into a preset region growing algorithm to perform quality grade classification and obtain a welding quality assessment report.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images