Industrial visual identification method and system based on deep learning

Abstract

The invention belongs to the technical field of quality detection, and discloses an industrial visual identification method and system based on deep learning. The method comprises the following steps: constructing a high-quality welding image sequence through multi-angle image acquisition, multi-scale noise reduction optimization and welding spot feature enhancement processing; adopting adaptive feature extraction and cross-scale feature fusion technologies to generate a fusion welding feature tensor; the attention mechanism enhancement and automatic segmentation technology is utilized to realize accurate extraction of a welding spot area and microdefect enhancement identification; combining historical welding quality evaluation data to perform dynamic characteristic correlation analysis, and constructing a quality evaluation model; establishing a welding quality diagnosis strategy through multi-dimensional defect type identification and self-adaptive quality grade evaluation; and finally, dynamic quality prediction and real-time monitoring of the welding spot defects are realized. Accurate evaluation and real-time monitoring of the welding quality are achieved, and the quality cost and the rework rate are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection, and more specifically, to an industrial vision recognition method and system based on deep learning. Background Art

[0002] With the rapid development of the electronics manufacturing industry and the intensification of the trend of product miniaturization, as a key process for connecting electronic components, wire bonding quality control has increasingly become a decisive factor affecting product reliability and performance. Traditional wire bonding quality inspection mainly relies on manual visual inspection, which is not only inefficient but also greatly affected by the experience and subjective judgment of inspectors, making it difficult to meet the requirements of modern electronics manufacturing for high-precision and high-consistency quality control.

[0003] In recent years, machine vision technology has been widely used in the field of industrial inspection, and automated welding quality inspection systems based on image processing have gradually replaced manual inspection. Currently, the common welding vision inspection systems on the market mainly use traditional image processing algorithms, such as edge detection, morphological analysis, and threshold segmentation, to identify and evaluate solder joints. These systems can achieve basic defect detection functions under ideal conditions, but face many challenges in the actual production environment.

[0004] Traditional vision inspection systems perform poorly in complex industrial environments. Especially when facing actual working conditions such as light changes, surface reflection, and environmental interference, the detection accuracy drops significantly, resulting in a large number of small defects being missed. On high-speed production lines, the inspection system cannot adapt to the subtle changes in the shape and position of solder joints in real time, resulting in a high misjudgment rate and having to rely on manual re-inspection, which greatly increases the labor cost and production cycle. Especially for the micro-solder joints of high-density electronic components, the existing technologies have low accuracy in identifying key quality problems such as false soldering, cold soldering, and insufficient solder joints, allowing a large number of potential quality hazards to flow into downstream processes or even end users, leading to product reliability risks and brand reputation losses. In addition, the existing systems lack the ability to deeply mine and learn historical data and cannot establish a correlation analysis between defects and process parameters, making quality problems occur frequently and repeatedly without being solved.

[0005] In view of this, the present invention proposes an industrial vision recognition method and system based on deep learning to solve the above problems. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: An industrial vision recognition method based on deep learning, comprising: Step S1: Obtain a multi-angle image dataset of wire bonding; perform multi-scale noise reduction optimization on the multi-angle image dataset of wire bonding, and perform solder joint feature enhancement processing to obtain an enhanced wire bonding image sequence; Step S2: Perform adaptive feature extraction on the enhanced core wire welding image sequence to generate a multi-level welding feature map; perform cross-scale feature fusion on the multi-level welding feature map to generate a fused welding feature tensor; Step S3: Automatically segment the solder joint area from the fused welding feature tensor to obtain a solder joint area feature matrix; perform microscopic defect enhancement recognition on the solder joint area feature matrix to obtain a solder joint defect feature space; Step S4: Obtain historical welding quality assessment data; perform dynamic feature correlation analysis on the historical welding quality assessment data to construct a welding quality assessment model; Step S5: Use the welding quality assessment model to perform multi-dimensional defect type recognition on the solder joint defect feature space and perform adaptive quality level assessment to construct a welding quality diagnosis strategy; Step S6: Based on the welding quality diagnosis strategy, perform dynamic quality prediction on the solder joint defect feature space to construct a real-time monitoring model for core wire welding quality.

[0007] Preferably, the specific steps of Step S1 are as follows: Step S11: Obtain a multi-angle image dataset of core wire welding; Step S12: Perform illumination unevenness correction on the multi-angle image dataset of core wire welding to obtain illumination equalized image data; Step S13: Perform multi-scale wavelet transform denoising on the illumination equalized image data to identify image noise interference regions; Step S14: Perform adaptive filtering on the illumination equalized image data according to the noise interference regions to obtain denoised image data; Step S15: Perform solder joint edge enhancement on the denoised image data to generate edge enhanced image data; Step S16: Perform contrast adaptive adjustment on the edge enhanced image data to obtain an enhanced core wire welding image sequence.

[0008] Preferably, the specific steps of Step S15 are as follows: Perform multi-directional gradient calculation on the denoised image data to obtain solder joint edge gradient features; Extract edge candidate regions according to the solder joint edge gradient features to obtain solder joint edge candidate regions; Perform edge continuity analysis on the solder joint edge candidate regions to generate an edge continuity score; Perform parameterized edge enhancement on the denoised image data based on the edge continuity score to generate an edge sharpening parameter range; Perform adaptive sharpening on the denoised image data based on the edge sharpening parameter range to mark key edge feature points; Perform non - linear edge enhancement optimization on key edge feature points to generate edge - enhanced image data.

[0009] Preferably, the specific steps of step S2 are as follows: Step S21: Perform multi - scale convolutional network processing on the enhanced core - wire welding image sequence to generate feature maps of different scales; Step S22: Perform solder joint feature recognition and analysis on feature maps of different scales to obtain multi - level solder joint feature representations; Step S23: Perform adaptive feature channel weighting based on the multi - level solder joint feature representations to generate multi - level welding feature maps; Step S24: Calculate the feature similarity of the multi - level welding feature maps to obtain a feature similarity matrix; Step S25: Perform feature - level difference standardization on the multi - level welding feature maps to generate a standardized multi - level feature map; Step S26: Perform cross - scale feature fusion on the standardized multi - level feature map based on the feature similarity matrix to generate a fused welding feature tensor.

[0010] Preferably, the specific steps of step S3 are as follows: Step S31: Enhance the fused welding feature tensor through an attention mechanism to obtain a solder joint attention region tensor; Step S32: Automatically segment the solder joint region of the solder joint attention region tensor to obtain a solder joint region feature matrix; Step S33: Perform high - frequency detail enhancement processing on the solder joint region feature matrix to obtain a microscopic detail - enhanced feature matrix; Step S34: Perform microscopic defect enhancement recognition on the microscopic detail - enhanced feature matrix to obtain a solder joint defect feature space.

[0011] Preferably, the specific steps of step S4 are as follows: Step S41: Obtain historical welding quality assessment data and corresponding image annotation data; Step S42: Analyze the distribution of defect types in the historical welding quality assessment data to generate a defect type distribution map; Step S43: Perform dynamic feature correlation analysis on the defect type distribution map to generate welding quality assessment features; Step S44: Perform multi - level neural network training on the welding quality assessment features to construct a welding quality assessment model.

[0012] Preferably, the specific steps of step S43 are as follows: Extract various defect feature parameters based on the defect type distribution map; Calculate the defect severity based on various defect characteristic parameters for the historical welding quality assessment data to obtain the defect severity index; Perform defect spatial distribution analysis on the defect type distribution map to obtain the defect spatial distribution characteristics; Perform correlation analysis on the defect spatial distribution characteristics according to the defect severity index to obtain the defect correlation characteristic data; Identify the influencing factors of welding process parameters based on the defect type distribution map; Perform defect cause correlation analysis based on the influencing factors of the process parameters to obtain the process parameter correlation data; Perform dynamic feature fusion analysis on the process parameter correlation data and the defect correlation characteristic data to generate the welding quality assessment features.

[0013] Preferably, the specific steps of step S5 are as follows: Step S51: Use the welding quality assessment model to perform multi-class defect recognition on the solder joint defect feature space and extract potential defect feature data; Step S52: Perform defect morphological analysis on the solder joint defect feature space to obtain the defect morphological feature link; Step S53: Perform multi-dimensional defect type recognition on the potential defect feature data based on the defect morphological feature link to obtain the defect type recognition result; Step S54: Quantify the quality influence degree of the defect type recognition result to generate the welding quality assessment value; Step S55: Perform adaptive quality level assessment based on the welding quality assessment value and construct a welding quality diagnosis strategy.

[0014] Preferably, the specific steps of step S6 are as follows: Step S61: Construct a quality prediction model for the solder joint defect feature space to generate a welding quality prediction framework; Step S62: Map the diagnosis rules to the welding quality prediction framework based on the welding quality diagnosis strategy to construct a core wire welding quality diagnosis framework; Step S63: Perform dynamic quality prediction on the core wire welding quality diagnosis framework to construct a core wire welding quality real-time monitoring model.

[0015] An industrial vision recognition system based on deep learning, which is used to implement the industrial vision recognition method based on deep learning, includes: A processing module, which is used to obtain a multi-angle image dataset of core wire welding; perform multi-scale noise reduction optimization on the multi-angle image dataset of core wire welding, and perform solder joint feature enhancement processing to obtain an enhanced core wire welding image sequence; A depth-level module for adaptively extracting features from the enhanced core wire welding image sequence to generate a multi-level welding feature map; performing cross-scale feature fusion on the multi-level welding feature map to generate a fused welding feature tensor; A region space module for automatically segmenting the solder joint area of the fused welding feature tensor to obtain a solder joint area feature matrix; performing microscopic defect enhancement recognition on the solder joint area feature matrix to obtain a solder joint defect feature space; A model construction module for obtaining historical welding quality assessment data; performing dynamic feature correlation analysis on the historical welding quality assessment data to construct a welding quality assessment model; A strategy fitting module for using the welding quality assessment model to perform multi-dimensional defect type recognition on the solder joint defect feature space and perform adaptive quality level assessment to construct a welding quality diagnosis strategy; A comprehensive module for dynamically predicting the quality of the solder joint defect feature space based on the welding quality diagnosis strategy to construct a real-time monitoring model for the core wire welding quality; the various modules are connected by wired and / or wireless means.

[0016] The technical effects and advantages of an industrial vision recognition method and system based on deep learning according to the present invention: The beneficial effects of the present invention lie in significantly improving the overall performance and practical value of core wire welding quality detection. It solves the pain points in traditional detection, greatly improves the accuracy and sensitivity of solder joint defect recognition, especially achieving a qualitative leap in the detection ability of microscopic defects. It realizes an all-round and multi-angle quality assessment system, which can not only accurately distinguish different types of defects, but also quantitatively evaluate the impact of defects on product quality. Thanks to the application of adaptive technology, it has strong adaptability and robustness to various complex working conditions and abnormal situations, effectively reducing the false judgment rate and missed detection rate. The real-time monitoring and early warning mechanism enables the production line to timely discover and respond to potential quality risks, prevent the generation of defective products from the source, and significantly reduce the quality cost and rework rate. In addition, by exploring the internal relationship between defects and process parameters, it provides data support and decision-making basis for process optimization and continuous quality improvement, ultimately achieving a double improvement in production efficiency and product quality. Brief Description of the Drawings

[0017] Figure 1 It is a schematic diagram of an industrial vision recognition method based on deep learning according to the present invention; Figure 2 It is a schematic diagram of an industrial vision recognition system based on deep learning according to the present invention. Detailed Embodiments

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] For ease of understanding, the specific process of the embodiments of this application will be described below. Please refer to Figure 1 One embodiment of the industrial vision recognition method based on deep learning in the embodiments of this application includes: Step S1: Obtain a multi-angle image dataset of core wire welding; perform multi-scale noise reduction optimization on the multi-angle image dataset of core wire welding, and perform solder joint feature enhancement processing to obtain an enhanced core wire welding image sequence; Step S2: Perform adaptive feature extraction on the enhanced core wire welding image sequence to generate a multi-level welding feature map; perform cross-scale feature fusion on the multi-level welding feature map to generate a fused welding feature tensor; Step S3: Automatically segment the solder joint area of the fused welding feature tensor to obtain a solder joint area feature matrix; perform microscopic defect enhancement recognition on the solder joint area feature matrix to obtain a solder joint defect feature space; Step S4: Obtain historical welding quality evaluation data; perform dynamic feature correlation analysis on the historical welding quality evaluation data to construct a welding quality evaluation model; Step S5: Use the welding quality evaluation model to perform multi-dimensional defect type recognition on the solder joint defect feature space, and perform adaptive quality level evaluation to construct a welding quality diagnosis strategy; Step S6: Perform dynamic quality prediction on the solder joint defect feature space based on the welding quality diagnosis strategy to construct a real-time monitoring model for the core wire welding quality.

[0020] It can be understood that the execution subject of this application can be an industrial vision recognition system based on deep learning, or a terminal or a server. Specifically, it is not limited here. The embodiments of this application will be described by taking the server as the execution subject as an example.

[0021] Specifically, obtain a multi-angle image dataset of core wire welding, including core wire welding images collected from different angles. Perform illumination non-uniformity correction on the multi-angle image dataset of core wire welding to obtain illumination equalized image data. The illumination non-uniformity correction uses methods such as histogram equalization or adaptive gamma correction to eliminate image quality problems caused by uneven illumination. Perform multi-scale wavelet transform denoising on the illumination equalized image data to identify image noise interference regions. The multi-scale wavelet transform can decompose the image into different scales and identify and process noise components at each scale. Perform adaptive filtering on the illumination equalized image data according to the noise interference regions to obtain denoised image data. The adaptive filtering dynamically adjusts the filtering parameters according to the noise characteristics, effectively removing noise while preserving image details. Perform solder joint edge enhancement on the denoised image data to generate edge enhanced image data. The solder joint edge enhancement improves the clarity and recognizability of the solder joint edges through multi-directional gradient calculation and edge continuity analysis. Perform contrast adaptive adjustment on the edge enhanced image data to obtain an enhanced core wire welding image sequence. The contrast adaptive adjustment dynamically adjusts the contrast parameters according to the image content to improve the distinguishability between the solder joint area and the background.

[0022] Perform multi-scale convolutional network processing on the enhanced core wire welding image sequence to generate feature maps of different scales. The multi-scale convolutional network captures multi-scale feature information of the solder joints through convolutional kernels of different sizes. Perform solder joint feature recognition and analysis on the feature maps of different scales to obtain multi-level solder joint feature representations. The solder joint feature recognition and analysis extracts features such as the shape, texture, and structure of the solder joints through a deep learning model. Perform adaptive feature channel weighting based on the multi-level solder joint feature representations to generate a multi-level welding feature map. The feature channel weighting assigns different weights according to the importance of each channel for solder joint recognition, highlighting key features. Calculate the feature similarity of the multi-level welding feature map to obtain a feature similarity matrix. The feature similarity calculation evaluates the correlation degree between features of different levels, providing a basis for subsequent fusion. Perform feature level difference normalization on the multi-level welding feature map to generate a normalized multi-level feature map. The feature level difference normalization eliminates the scale differences of features at different levels, making feature fusion more effective. Perform cross-scale feature fusion on the normalized multi-level feature map based on the feature similarity matrix to generate a fused welding feature tensor. The cross-scale feature fusion integrates multi-level feature information to form a more comprehensive welding feature representation.

[0023] Enhance the fusion welding feature tensor with an attention mechanism to obtain the solder joint attention region tensor. The attention mechanism enhancement guides the model to focus on the key regions of the solder joints by learning the importance weights of the solder joint regions. Automatically segment the solder joint region from the solder joint attention region tensor to obtain the solder joint region feature matrix. The automatic segmentation of the solder joint region accurately locates the solder joint region through a semantic segmentation network, excluding background interference. Perform high-frequency detail enhancement processing on the solder joint region feature matrix to obtain the micro-detail enhanced feature matrix. The high-frequency detail enhancement processing highlights the microstructural features of the solder joints, facilitating subsequent defect identification. Perform micro-defect enhanced identification on the micro-detail enhanced feature matrix to obtain the solder joint defect feature space. The micro-defect enhanced identification highlights the potential defect features through a feature enhancement network, forming a defect feature space representation.

[0024] Obtain historical welding quality assessment data and corresponding image annotation data, including expert evaluation results and defect annotation information. Analyze the defect type distribution of the historical welding quality assessment data to generate a defect type distribution map. The defect type distribution analysis statistically analyzes the occurrence frequencies and distribution characteristics of various defects to form a defect distribution map. Conduct dynamic feature correlation analysis on the defect type distribution map to generate welding quality assessment features. The dynamic feature correlation analysis explores the correlation between defect features and welding quality, extracting key features for quality assessment. Train a multi-layer neural network with the welding quality assessment features to construct a welding quality assessment model. The multi-layer neural network training establishes a mapping relationship between defect features and quality assessment through deep learning algorithms to form a quality assessment model.

[0025] Use the welding quality assessment model to perform multi-class defect identification on the solder joint defect feature space and extract potential defect feature data. The multi-class defect identification identifies various welding defects, including pores, cracks, lack of fusion, etc., through a trained deep learning model. Conduct defect morphology analysis on the solder joint defect feature space to obtain a defect morphology feature link. The defect morphology analysis studies the geometric shapes, sizes, and distribution characteristics of defects to construct a defect morphology feature link. Based on the defect morphology feature link, perform multi-dimensional defect type identification on the potential defect feature data to obtain defect type identification results. The multi-dimensional defect type identification comprehensively considers factors such as the morphology, location, and severity of defects to accurately judge the defect type. Quantify the degree of quality impact of the defect type identification results to generate a welding quality assessment value. The quantification of the degree of quality impact calculates the comprehensive quality assessment value based on the impact weights of different types of defects on welding quality. Based on the welding quality assessment value, perform adaptive quality level assessment to construct a welding quality diagnosis strategy. The adaptive quality level assessment maps the quality assessment value to different quality levels according to industry standards and specific application requirements and forms corresponding diagnosis strategies.

[0026] Construct a quality prediction model for the solder joint defect feature space to generate a welding quality prediction framework. The construction of the quality prediction model uses a time series deep learning network to establish a mapping relationship between defect features and quality change trends. Based on the welding quality diagnosis strategy, map the diagnosis rules to the welding quality prediction framework to construct a core wire welding quality diagnosis framework. The diagnosis rule mapping transforms the quality diagnosis strategy into an executable rule set to guide real-time quality monitoring. Perform dynamic quality prediction on the core wire welding quality diagnosis framework to construct a real-time monitoring model for core wire welding quality. The dynamic quality prediction continuously analyzes the changes in defect features during the welding process to achieve real-time monitoring and early warning of welding quality.

[0027] In the embodiments of the present application, through multi-scale noise reduction optimization and solder joint feature enhancement processing, the quality of the original image is effectively improved, laying a foundation for subsequent analysis. Through adaptive feature extraction and cross-scale feature fusion, comprehensive capture and effective integration of solder joint features are achieved. Through automatic segmentation of the solder joint area and enhanced recognition of microscopic defects, the solder joint area is accurately located and potential defect features are highlighted. Through dynamic feature correlation analysis and multi-level neural network training, an accurate and reliable welding quality evaluation model is constructed. Through multi-dimensional defect type recognition and adaptive quality level evaluation, accurate diagnosis and grading of welding quality are achieved. Through dynamic quality prediction and construction of a real-time monitoring model, continuous monitoring and early warning of welding quality are realized, improving production efficiency and product quality.

[0028] In a specific embodiment, the process of executing step S1 specifically includes the following steps: Step S11: Obtain a multi-angle image dataset of core wire welding; Step S12: Perform illumination non-uniformity correction on the multi-angle image dataset of core wire welding to obtain illumination equalized image data; Step S13: Perform multi-scale wavelet transform denoising on the illumination equalized image data to identify image noise interference regions; Step S14: Perform adaptive filtering on the illumination equalized image data according to the noise interference regions to obtain denoised image data; Step S15: Perform solder joint edge enhancement on the denoised image data to generate edge enhanced image data; Step S16: Perform contrast adaptive adjustment on the edge enhanced image data to obtain an enhanced core wire welding image sequence.

[0029] Specifically, obtain a multi-angle image dataset of core wire welding. Collect core wire welding images from different angles through an industrial camera array to ensure that the solder joint features are captured comprehensively. Multi-angle collection helps reduce the occlusion problem of a single perspective and improve the integrity of feature extraction. Perform illumination unevenness correction on the multi-angle image dataset of core wire welding to obtain illumination equalized image data. Illumination unevenness correction adjusts the pixel intensity distribution of the image through histogram equalization technology to improve the image contrast. For example, for image I(x, y), its histogram equalization is expressed as: I_eq(x, y)=T[I(x, y)]; where T[] is the transformation function, x and y are the coordinate indices of the pixels, mapping the gray values of the original image to a new gray value range. Perform multi-scale wavelet transform denoising on the illumination equalized image data to identify the image noise interference regions. The multi-scale wavelet transform decomposes the image into different frequency components, expressed as: ; where D_j(x, y) represents the detail component at scale j, and A_J(x, y) represents the approximation component at scale J. By thresholding the wavelet coefficients at each scale, the noise can be effectively identified and suppressed. Perform adaptive filtering on the illumination equalized image data according to the noise interference regions to obtain denoised image data. Adaptive filtering dynamically adjusts the filtering parameters according to the local noise characteristics, expressed as: I_d(x, y)=I_e(x, y)×h(x, y, σ(x, y)); where h() is the filter kernel function, and σ(x, y) is the parameter adaptively adjusted according to the local noise level. Perform solder joint edge enhancement on the denoised image data to generate edge enhanced image data. The solder joint edge enhancement process includes steps such as multi-directional gradient calculation, edge candidate region extraction, and edge continuity analysis, effectively improving the clarity of the solder joint edges. Perform contrast adaptive adjustment on the edge enhanced image data to obtain an enhanced core wire welding image sequence. Contrast adaptive adjustment dynamically adjusts the contrast parameters according to the image content, expressed as: I_e(x, y)=α(x, y)×I_b(x, y)+β(x, y); where α(x, y) and β(x, y) are the gain and bias parameters adaptively adjusted according to the local image features, I_b(x, y) is the image after edge enhancement, and I_e(x, y) is the final enhanced image. Through this series of processes, an enhanced core wire welding image sequence with significantly improved quality is obtained, providing a high-quality data basis for subsequent analysis.

[0030] In a specific embodiment, the process of performing step S15 specifically includes the following steps: Perform multi-directional gradient calculation on the denoised image data to obtain the solder joint edge gradient features; Extract the edge candidate region according to the solder joint edge gradient feature to obtain the solder joint edge candidate region; Perform edge continuity analysis on the solder joint edge candidate region to generate an edge continuity score; Perform parametric edge enhancement processing on the denoised image data based on the edge continuity score to generate an edge sharpening parameter range; Perform adaptive sharpening processing on the denoised image data based on the edge sharpening parameter range to mark the key edge feature points; Perform non-linear edge enhancement optimization on the key edge feature points to generate edge enhanced image data.

[0031] Specifically, perform multi-directional gradient calculation on the denoised image data to obtain the solder joint edge gradient feature. The multi-directional gradient calculation calculates the image gradient in different directions through operators such as Sobel, Prewitt, or Canny. Through the multi-directional gradient calculation, the directional features of the solder joint edge can be comprehensively captured. Extract the edge candidate region according to the solder joint edge gradient feature to obtain the solder joint edge candidate region. The edge candidate region extraction is performed by gradient magnitude threshold processing, and the region with a gradient magnitude greater than the threshold is selected as the edge candidate region. Perform edge continuity analysis on the edge candidate region of the solder joint to generate an edge continuity score. The edge continuity analysis evaluates the continuity degree of the edge by calculating the connectivity and curvature characteristics of the edge candidate region, expressed as: C(i)=f(L(i), K(i)); where L(i) is the length of the i-th edge region, K(i) is its curvature characteristic, f() is the scoring function, and C(i) is the continuity score of the i-th edge region.

[0032] ; where L(i) is the length of the i-th edge region, L_max is the maximum length among all edge regions, K_var(i) is the degree of curvature change (curvature variance or volatility) of the i-th edge region, K_max is the maximum curvature change among all edge regions, and w1 and w2 are weight coefficients, usually w1 + w2 = 1.

[0033] Perform parametric edge enhancement processing on the denoised image data based on the edge continuity score to generate an edge sharpening parameter range. The parametric edge enhancement processing adaptively determines the enhancement parameter range according to the continuity score, expressed as: P_range(i)=[P_min + α×C(i), P_max - β×(1 - C(i))]; where P_range(i) is the sharpening parameter range of the i-th edge region, α and β are adjustment parameters, and the preset range interval is [P_min, P_max]. Adaptive sharpening processing is performed on the denoised image data based on the edge sharpening parameter range, and key edge feature points are marked. The adaptive sharpening processing performs non-uniform sharpening on the image according to the sharpening parameter range to highlight the key edge features, expressed as: I_s(x, y)=I_d(x, y)+λ(x, y)×(I_d(x, y)-I_d(x, y)×h_blur(x, y)); where I_s(x, y) is the image after adaptive sharpening processing, h_blur is the blur kernel, and λ(x, y) is the local sharpening intensity determined according to the sharpening parameter range, which varies with position. The entire formula represents adding the extracted edge details back to the original image according to the local sharpening intensity, thereby enhancing the edge features. Nonlinear edge enhancement optimization is performed on the key edge feature points to generate edge-enhanced image data. The nonlinear edge enhancement optimization further enhances the edge features through a nonlinear mapping function, expressed as: I_b(x, y)=g(I_s(x, y)); where g() is a nonlinear mapping function (such as the Sigmoid function), and I_b(x, y) is the final edge-enhanced image data.

[0034] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Step S21: Perform multi-scale convolutional network processing on the enhanced core wire welding image sequence to generate feature maps of different scales; Step S22: Perform solder joint feature recognition and analysis on the feature maps of different scales to obtain multi-level solder joint feature representations; Step S23: Perform adaptive feature channel weighting based on the multi-level solder joint feature representations to generate a multi-level welding feature map; Step S24: Calculate the feature similarity of the multi-level welding feature map to obtain a feature similarity matrix; Step S25: Perform feature level difference normalization on the multi-level welding feature map to generate a normalized multi-level feature map; Step S26: Perform cross-scale feature fusion on the normalized multi-level feature map based on the feature similarity matrix to generate a fused welding feature tensor.

[0035] Specifically, perform multi-scale convolutional network processing on the enhanced core wire welding image sequence to generate feature maps of different scales. The multi-scale convolutional network captures the multi-scale feature information of the solder joints through convolutional kernels and pooling operations of different sizes, expressed as: $F_l = \text{CNN}_l(I_e)$; where $\text{CNN}_l$ represents the $l$-th layer convolutional network, and $F_l$ is the generated feature map. Through a multi-layer convolutional network, multi-scale feature representations from low-level textures to high-level semantics can be obtained. Feature recognition and analysis of solder joints are performed on feature maps of different scales to obtain multi-level solder joint feature representations. Feature recognition and analysis of solder joints extract features such as the shape, texture, and structure of solder joints through a feature enhancement network, which can be expressed as: $F_{l\_e}=\text{FEN}(F_l)$; where $\text{FEN}$ is the feature enhancement network used to strengthen and refine the features extracted by the original convolutional network. It includes a residual connection module (enhancing feature propagation), an attention module (highlighting important regions), dilated convolution (expanding the receptive field), and feature recalibration (adjusting the feature distribution); $F_{l\_e}$ is the enhanced feature representation. Adaptive feature channel weighting is performed based on the multi-level solder joint feature representation to generate a multi-level welding feature map. Feature channel weighting assigns different weights to feature channels through an attention mechanism to highlight key features, which is expressed as: $F_{l\_w}=F_{l\_e}\times W_l$; where $W_l$ is the channel weight learned through the attention mechanism, which is a vector, and each element corresponds to the importance weight of a feature channel. $F_{l\_w}$ is the feature map after weighting for the $l$-th layer. Feature similarity calculation is performed on the multi-level welding feature map to obtain a feature similarity matrix. Feature similarity calculation evaluates the degree of association between features at different levels and is measured using cosine similarity or Pearson correlation coefficient. Feature level difference normalization is performed on the multi-level welding feature map to generate a normalized multi-level feature map. Feature level difference normalization eliminates the scale differences of features at different levels, which is expressed as: ; where $\mu_l$ and $\sigma_l$ are the mean and standard deviation of the features for the $l$-th layer respectively, and $F_{l\_n}$ is the normalized feature map, eliminating the scale differences between features at different levels. Cross-scale feature fusion is performed on the normalized multi-level feature map based on the feature similarity matrix to generate a fused welding feature tensor. Cross-scale feature fusion integrates multi-level feature information, which is expressed as: ; where $w_l$ is the fusion weight calculated based on the similarity matrix, which determines the importance of the features for the $l$-th layer in the final fusion result, $\text{up}()$ is the upsampling operation to align feature maps of different scales, and $F_{\text{fusion}}$ is the final fused feature tensor.

[0036] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Step S31: Perform attention mechanism enhancement on the fused welding feature tensor to obtain a solder joint attention region tensor; Step S32: Perform automatic segmentation of the solder joint region on the solder joint attention region tensor to obtain a solder joint region feature matrix; Step S33: Perform high-frequency detail enhancement processing on the solder joint area feature matrix to obtain a micro-detail enhanced feature matrix; Step S34: Perform micro-defect enhanced recognition on the micro-detail enhanced feature matrix to obtain a solder joint defect feature space.

[0037] Specifically, perform attention mechanism enhancement on the fused welding feature tensor to obtain a solder joint attention region tensor. The attention mechanism enhancement combines spatial attention and channel attention to highlight the key area of the solder joint, expressed as: A_spatial = σ(f_s(F_fusion)); A_channel = σ(f_c(F_fusion)); F_att = F_fusion ⊙ A_spatial ⊙ A_channel; where A_spatial is the spatial attention map, representing the importance weight of each spatial position on the feature map; f_s is the spatial attention calculation function, usually calculating the importance of each position on the feature map through convolution operations, A_channel is the channel attention vector, representing the importance weight of each feature channel; f_c is the channel attention calculation function, usually calculating the importance of each feature channel through global pooling and fully connected layers, σ is the activation function, usually the Sigmoid function, mapping the value to the range of 0 - 1, and F_att is the feature tensor after attention enhancement. Perform automatic segmentation of the solder joint area on the solder joint attention region tensor to obtain the solder joint area feature matrix. The automatic segmentation of the solder joint area accurately locates the solder joint area through a semantic segmentation network, expressed as: M = Seg(F_att); F_roi = F_att ⊙ M; where Seg is a segmentation network, such as U-Net, DeepLab, etc., used to segment the image into different semantic regions, M is the segmentation mask, which is a binary image, the area with a value of 1 represents the solder joint, and the area with a value of 0 represents the background, ⊙ represents element-wise multiplication, and F_roi is the solder joint area feature matrix, which only retains the features of the solder joint area and excludes background interference. Perform high-frequency detail enhancement processing on the solder joint area feature matrix to obtain a micro-detail enhanced feature matrix. The high-frequency detail enhancement processing highlights the microstructural features of the solder joint, expressed as: F_detail = F_roi + λ1 × HF(F_roi); where HF is a high-frequency extraction operation, which can be a high-pass filter, Laplacian operator, wavelet transform, etc., for extracting the high-frequency details of the image. λ1 is an enhancement coefficient that controls the intensity of high-frequency detail enhancement. F_detail is the feature matrix after detail enhancement, highlighting the microscopic structure features of the solder joints. Micro-defect enhancement recognition is performed on the micro-detail enhanced feature matrix to obtain the solder joint defect feature space. Micro-defect enhancement recognition highlights potential defect features through a feature enhancement network, expressed as: F_defect = DefectNet(F_detail); where DefectNet is a defect enhancement network, a neural network for identifying and enhancing solder joint defect features, including a residual module (to improve feature extraction ability), an attention mechanism (to focus on potential defect areas), multi-scale feature fusion (to capture defect features at different scales), and a contrast learning module (to enhance the difference between defects and normal areas). F_defect is the final solder joint defect feature space, which is the feature representation ultimately used for defect analysis and quality assessment.

[0038] In a specific embodiment, the process of executing step S4 specifically includes the following steps: Step S41: Obtain historical welding quality assessment data and corresponding image annotation data; Step S42: Conduct defect type distribution analysis on the historical welding quality assessment data to generate a defect type distribution map; Step S43: Conduct dynamic feature correlation analysis on the defect type distribution map to generate welding quality assessment features; Step S44: Conduct multi-level neural network training on the welding quality assessment features to construct a welding quality assessment model.

[0039] Specifically, obtain historical welding quality assessment data and corresponding image annotation data. The historical data includes expert evaluation results, defect annotation information, and relevant process parameters, providing labeled samples for model training. Conduct defect type distribution analysis on the historical welding quality assessment data to generate a defect type distribution map. Defect type distribution analysis statistically analyzes the occurrence frequency and distribution characteristics of various defects, expressed as: ; ; where, defect_i is the i-th type of defect, such as specific defect types like pores, cracks, lack of fusion, etc., count(defect_i) is the number of times the i-th type of defect appears in the historical data, total is the total number of samples in the historical data, D(i) is the frequency distribution of the i-th type of defect, representing the probability of this type of defect appearing in all samples, P(i, j) is the conditional probability distribution of the i-th type of defect at the j-th position, representing the probability of this type of defect appearing at a specific position, location_j is the j-th position area, such as the center of the solder joint, the edge, the heat affected zone, etc., and count(defect_i, location_j) is the number of times the i-th type of defect appears in the j-th position area. Perform dynamic feature correlation analysis on the defect type distribution map to generate welding quality evaluation features. The dynamic feature correlation analysis explores the correlation between defect features and welding quality, expressed as: R(i, q) = corr(defect_i, quality_q); F_quality = fl({R(i, q)}, {D(i)}, {P(i, j)}); where R(i, q) is the correlation between the i-th type of defect and the quality index q, representing the degree of influence of the defect on the quality, quality_q is the q-th quality index, such as strength, durability, conductivity, etc.; corr() is the correlation calculation function, such as the Pearson correlation coefficient, the Spearman rank correlation coefficient, etc., F_quality is the extracted quality evaluation feature, which synthesizes information on defect types, distributions, and quality impacts, {R(i, q)} is the set of correlation coefficients between all defect types and quality indices, {D(i)} is the set of frequency distributions of all defect types, {P(i, j)} is the set of conditional probability distributions of all defect types at each position, and fl() is the feature fusion function, which may be a weighted combination, a non-linear transformation, or a neural network, etc. Perform multi-level neural network training on the welding quality evaluation features to construct a welding quality evaluation model. The multi-level neural network training establishes a mapping relationship between defect features and quality evaluation through a deep learning algorithm, expressed as: Q = DNN(F_quality); where DNN is a deep neural network, including multiple hidden layers, convolutional layers, fully connected layers, etc., for learning the complex mapping relationship between defect features and quality evaluation, and Q is the predicted quality evaluation result, which is the quality grade, the qualified rate, or a specific quality score.

[0040] In a specific embodiment, the process of performing step S43 may specifically include the following steps: Extract various defect feature parameters based on the defect type distribution map; Calculate the defect severity based on the various defect feature parameters for the historical welding quality evaluation data to obtain a defect severity index; Perform defect spatial distribution analysis on the defect type distribution map to obtain defect spatial distribution characteristics; Perform correlation analysis on the defect spatial distribution characteristics according to the defect severity index to obtain defect correlation characteristic data; Identify the influencing factors of welding process parameters based on the defect type distribution map; Perform defect cause correlation analysis based on the influencing factors of the process parameters to obtain process parameter correlation data; Perform dynamic feature fusion analysis on the process parameter correlation data and the defect correlation characteristic data to generate welding quality evaluation characteristics.

[0041] Specifically, extract various defect feature parameters based on the defect type distribution map. The defect feature parameters include characteristics such as the size, shape, density, and distribution of the defects, and can be expressed as: P_size(i)=avg(defect_i); P_shape(i)=shape(defect_i); P_density(i)=density(defect_i); where P_size(i), P_shape(i), and P_density(i) respectively represent the size, shape, and density parameters of the i-th type of defect, avg() is the average size calculation function, based on area, diameter, or length, shape() is the shape factor calculation function, which can be circularity, aspect ratio, or complexity, etc., and density() is the density calculation function, representing the number of defects per unit area. Calculate the defect severity of the historical welding quality evaluation data according to various defect feature parameters to obtain the defect severity index. The defect severity calculation comprehensively considers the feature parameters and location of the defects, and is expressed as: SL(i)=w_s×P_size(i)+w_p×P_shape(i)+w_d×P_density(i)+w_loc×P_loc(i); where w_s, w_p, w_d, and w_loc are the weights of each parameter, P_loc(i) is the location parameter of the i-th type of defect, reflecting the importance of the defect location; SL(i) is the severity index of the i-th type of defect. Perform defect spatial distribution analysis on the defect type distribution map to obtain defect spatial distribution characteristics. The defect spatial distribution analysis studies the spatial distribution law of defects in the solder joint area, and is expressed as: D_spatial(x, y)=∑ i D(i)×P(i, x, y); Among them, D_spatial(x, y) represents the defect distribution density at the position (x, y), which represents the comprehensive probability of defects occurring at this position, and P(i, x, y) is the probability of the i-th type of defect at the position (x, y). According to the defect severity index, a correlation analysis is carried out on the spatial distribution characteristics of defects to obtain defect correlation feature data. The correlation analysis studies the relationship between defect severity and spatial distribution, which is expressed as: CL(i, j) = corr(SL(i), D_spatial_j); F_corr = {CL(i, j)}; where CL(i, j) is the correlation between the severity of the i-th type of defect and the distribution in the j-th region, D_spatial_j is the defect distribution density in the j-th region, corr() is the correlation calculation function, such as the Pearson correlation coefficient, and F_corr is the defect correlation feature data, which contains the correlation information between all defect types and regional distributions. Based on the defect type distribution map, the influencing factors of welding process parameters are identified. The influencing factors of process parameters include welding current, voltage, speed, shielding gas, etc. By analyzing the relationship between these parameters and defect distribution, the key influencing factors are identified. Based on the influencing factors of the process parameters, a defect cause correlation analysis is carried out to obtain process parameter correlation data. The defect cause correlation analysis studies the causal relationship between process parameters and defect formation, which is expressed as: R_param(s, j) = corr(param_s, defect_j); F_param = {R_param(s, j)}; where R_param(s, j) is the correlation between the s-th process parameter and the j-th type of defect, param_s is the s-th process parameter, such as welding current, voltage, speed, shielding gas, etc., and F_param is the process parameter correlation data, which contains the correlation information between all process parameters and defect types. A dynamic feature fusion analysis is carried out on the process parameter correlation data and the defect correlation feature data to generate welding quality evaluation features. The dynamic feature fusion analysis integrates the correlation information of process parameters and defect features, which can be expressed as: F_quality = Fusion(F_corr, F_param); where Fusion is the feature fusion function, which can be weighted combination, non-linear transformation or neural network, etc., and F_quality is the final welding quality evaluation feature, which synthesizes the correlation information of defect features and process parameters.

[0042] In a specific embodiment, the process of executing step S5 may specifically include the following steps: Step S51: Use the welding quality evaluation model to perform multi-class defect identification on the solder joint defect feature space and extract potential defect feature data; Step S52: Conduct defect morphological analysis on the solder joint defect feature space to obtain a defect morphological feature link; Step S53: Perform multi-dimensional defect type recognition on the potential defect feature data based on the defect morphological feature link to obtain a defect type recognition result; Step S54: Quantify the degree of quality impact of the defect type recognition result to generate a welding quality evaluation value; Step S55: Conduct adaptive quality level evaluation based on the welding quality evaluation value to construct a welding quality diagnosis strategy.

[0043] Specifically, use a welding quality evaluation model to perform multi-category defect recognition on the solder joint defect feature space and extract potential defect feature data. Multi-category defect recognition uses a trained deep learning model to recognize various welding defects, which can be expressed as: P(c_i|F_defect)=QualityModel(F_defect); F_potential=Extract(F_defect, P); where P(c_i|F_defect) is the probability that the defect feature space belongs to the i-th type of defect, and F_potential is the extracted potential defect feature data. Conduct defect morphological analysis on the solder joint defect feature space to obtain a defect morphological feature link. Defect morphological analysis studies the geometric shape and topological characteristics of defects, which can be expressed as: M_shape(i)=ShapeAnalysis(F_defect, c_i); L_morph={M_shape(i)}; where M_shape(i) is the morphological feature of the i-th type of defect, and L_morph is the morphological feature link. Perform multi-dimensional defect type recognition on the potential defect feature data based on the defect morphological feature link to obtain a defect type recognition result. Multi-dimensional defect type recognition comprehensively considers factors such as the morphology, location, and severity of defects, which can be expressed as: R_defect=Classify(F_potential, L_morph); where Classify is a multi-dimensional classification function, and R_defect is the defect type recognition result. Quantify the degree of quality impact of the defect type recognition result to generate a welding quality evaluation value. The quantification of the degree of quality impact calculates a comprehensive quality evaluation value according to the impact weights of different types of defects on welding quality, which can be expressed as: Q_value=∑ i[w_i×count(defect_i)×severity(defect_i)]; where w_i is the weight coefficient of the i-th type of defect, count(defect_i) is the quantity of this type of defect, severity(defect_i) is its severity, and Q_value is the comprehensive quality evaluation value. Based on the welding quality evaluation value, an adaptive quality grade evaluation is performed to construct a welding quality diagnosis strategy. The adaptive quality grade evaluation maps the quality evaluation value to different quality grades according to industry standards and specific application requirements, which can be expressed as: Grade = Mapping(Q_value); Strategy = DiagnosisRule(Grade, R_defect); where Mapping is the quality grade mapping function, DiagnosisRule is the diagnostic rule generation function, Grade is the quality grade, and Strategy is the welding quality diagnosis strategy.

[0044] In a specific embodiment, the process of executing step S6 specifically includes the following steps: Step S61: Construct a quality prediction model for the solder joint defect feature space to generate a welding quality prediction framework; Step S62: Perform diagnostic rule mapping on the welding quality prediction framework based on the welding quality diagnosis strategy to construct a core wire welding quality diagnosis framework; Step S63: Perform dynamic quality prediction on the core wire welding quality diagnosis framework to construct a real-time monitoring model for core wire welding quality.

[0045] Specifically, construct a quality prediction model for the solder joint defect feature space to generate a welding quality prediction framework. The construction of the quality prediction model establishes a mapping relationship between defect features and quality change trends through a time series deep learning network, which can be expressed as: Q_t+1 = fP(F_defect_t, F_defect_t-1,..., F_defect_t-n); Frame = {fP, para}; where Q_t+1 is the predicted welding quality value at time t+1, representing the welding quality state at a future time. fP is a prediction function, usually implemented using time series networks such as LSTM or GRU. F_defect_t is the solder joint defect feature at time t, the defect data at the current time, and para are the training parameters of the prediction model, including network weights, biases, etc.; Frame is the quality prediction framework. Based on the welding quality diagnosis strategy, a diagnostic rule mapping is performed on the welding quality prediction framework to construct a core wire welding quality diagnosis framework. The diagnostic rule mapping transforms the quality diagnosis strategy into an executable rule set, transforms the abstract diagnosis strategy into specific rules, and integrates the prediction framework and the rule set into a diagnosis framework. The quality diagnosis framework integrates the prediction framework and diagnostic rules. Dynamic quality prediction is performed on the core wire welding quality diagnosis framework to construct a real-time monitoring model for core wire welding quality. The dynamic quality prediction continuously analyzes the change in defect features during the welding process to achieve real-time monitoring of welding quality.

[0046] The above describes the industrial vision recognition method based on deep learning in the embodiments of the present application. Next, the industrial vision recognition system based on deep learning in the embodiments of the present application will be described. Please refer to Figure 2 , an embodiment of the industrial vision recognition system based on deep learning in the embodiments of the present application includes: A processing module, configured to obtain a multi-angle image dataset of core wire welding; perform multi-scale noise reduction optimization on the multi-angle image dataset of core wire welding, and perform solder joint feature enhancement processing to obtain an enhanced core wire welding image sequence; A deep hierarchical module, configured to perform adaptive feature extraction on the enhanced core wire welding image sequence to generate a multi-level welding feature map; perform cross-scale feature fusion on the multi-level welding feature map to generate a fused welding feature tensor; A region space module, configured to automatically segment the solder joint area of the fused welding feature tensor to obtain a solder joint area feature matrix; perform microscopic defect enhancement recognition on the solder joint area feature matrix to obtain a solder joint defect feature space; A model construction module, configured to obtain historical welding quality evaluation data; perform dynamic feature correlation analysis on the historical welding quality evaluation data to construct a welding quality evaluation model; A strategy fitting module, configured to use the welding quality evaluation model to perform multi-dimensional defect type recognition on the solder joint defect feature space, and perform adaptive quality level evaluation to construct a welding quality diagnosis strategy; A comprehensive module, based on the welding quality diagnosis strategy, performs dynamic quality prediction on the solder joint defect feature space to construct a real-time monitoring model for core wire welding quality; each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0047] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0048] In the embodiments of the present application, through multi-scale noise reduction optimization and solder joint feature enhancement processing, the quality of the original image is effectively improved, laying a foundation for subsequent analysis. The multi-scale noise reduction optimization adopts a method combining wavelet transform and adaptive filtering, which can effectively remove noises of different frequencies while retaining the detailed information of the solder joints. The solder joint feature enhancement processing highlights the key structural features of the solder joints through edge enhancement and contrast adjustment, improving the accuracy of subsequent analysis.

[0049] Through adaptive feature extraction and cross-scale feature fusion, the comprehensive capture and effective integration of solder joint features are achieved. The adaptive feature extraction adopts a multi-scale convolutional network and a feature channel weighting mechanism, which can dynamically adjust the feature extraction strategy according to the image content and capture the multi-level features of the solder joints. The cross-scale feature fusion effectively integrates the feature information of different scales through feature similarity calculation and hierarchical difference normalization, forming a more comprehensive representation of welding features.

[0050] Through automatic segmentation of the solder joint area and enhanced recognition of micro-defects, the solder joint area is accurately located and the potential defect features are highlighted. The automatic segmentation of the solder joint area adopts a method combining an attention mechanism and a semantic segmentation network, which can accurately locate the solder joint area and exclude background interference. The enhanced recognition of micro-defects highlights the potential defects in the microstructure of the solder joints through high-frequency detail enhancement and defect feature extraction, providing key information for subsequent defect analysis.

[0051] Through dynamic feature correlation analysis and multi-level neural network training, an accurate and reliable welding quality assessment model is constructed. The dynamic feature correlation analysis comprehensively considers the correlation relationships among defect features, spatial distributions, and process parameters, and extracts the key features for quality assessment. The multi-level neural network training establishes the mapping relationship between features and quality through deep learning algorithms to achieve accurate assessment of welding quality.

[0052] Through multi-dimensional defect type recognition and adaptive quality level assessment, the precise diagnosis and grading of welding quality are achieved. The multi-dimensional defect type recognition comprehensively considers the morphology, location, and severity of defects to accurately identify various welding defects. The adaptive quality level assessment maps the quality assessment value to a suitable quality level according to industry standards and application requirements and forms corresponding diagnostic strategies.

[0053] Through the construction of a dynamic quality prediction and real-time monitoring model, continuous monitoring and early warning of welding quality are achieved, improving production efficiency and product quality. The dynamic quality prediction uses a time series deep learning network, which can predict the change trend of welding quality and detect potential problems in a timely manner. The real-time monitoring model realizes real-time monitoring and early warning of welding quality through diagnostic rule mapping and continuous update, ensuring the stability of the production process and the consistency of product quality.

[0054] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0055] It should be noted that in this text, the terms "including", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0056] In the description of the present invention, it should be understood that the terms "first", "second", etc. are only used for descriptive distinction and cannot be understood as indicating or implying relative importance.

[0057] In the description of the present invention, unless otherwise specified, the meaning of "a plurality" is two or more.

[0058] In the description of the present invention, the meaning of "several" is one or more, and the meaning of "a large number" is two or more.

[0059] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0060] For the formulas in this specification, only the numerical values are calculated after dimensionlessization. The formulas are obtained by collecting a large amount of data and performing software simulations to get a formula that is closest to the actual situation. The preset parameters and threshold values in the formulas are set by those skilled in the art according to the actual situation.

[0061] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. An industrial vision recognition method based on deep learning, characterized in that, Including: Step S1: Obtain a multi-angle image dataset of core wire welding; Perform multi-scale noise reduction optimization on the multi-angle image dataset of core wire welding, and perform solder joint feature enhancement processing to obtain an enhanced core wire welding image sequence; Step S2: Perform adaptive feature extraction on the enhanced core wire welding image sequence to generate a multi-level welding feature map; perform cross-scale feature fusion on the multi-level welding feature map to generate a fused welding feature tensor; Step S3: Automatically segment the solder joint area of the fused welding feature tensor to obtain a solder joint area feature matrix; perform microscopic defect enhancement recognition on the solder joint area feature matrix to obtain a solder joint defect feature space; Step S4: Obtain historical welding quality assessment data; perform dynamic feature correlation analysis on the historical welding quality assessment data to construct a welding quality assessment model; Step S5: Use the welding quality assessment model to perform multi-dimensional defect type recognition on the solder joint defect feature space, and perform adaptive quality level assessment to construct a welding quality diagnosis strategy; Step S6: Based on the welding quality diagnosis strategy, perform dynamic quality prediction on the solder joint defect feature space to construct a real-time monitoring model for core wire welding quality.

2. The industrial vision recognition method based on deep learning according to claim 1, wherein The specific steps of Step S1 are as follows: Step S11: Obtain a multi-angle image dataset of core wire welding; Step S12: Perform illumination imbalance correction on the multi-angle image dataset of core wire welding to obtain illumination-equalized image data; Step S13: Perform multi-scale wavelet transform noise reduction on the illumination-equalized image data to identify image noise interference regions; Step S14: Perform adaptive filtering processing on the illumination-equalized image data according to the noise interference regions to obtain noise-reduced image data; Step S15: Perform solder joint edge enhancement processing on the noise-reduced image data to generate edge-enhanced image data; Step S16: Perform contrast adaptive adjustment processing on the edge-enhanced image data to obtain an enhanced core wire welding image sequence.

3. The industrial vision recognition method based on deep learning according to claim 2, wherein The specific steps of Step S15 are as follows: Perform multi-directional gradient calculation on the noise-reduced image data to obtain solder joint edge gradient features; Extract edge candidate regions according to the solder joint edge gradient features to obtain solder joint edge candidate regions; Perform edge continuity analysis on the solder joint edge candidate regions to generate an edge continuity score; Perform parametric edge enhancement processing on the noise-reduced image data based on the edge continuity score to generate an edge sharpening parameter range; Perform adaptive sharpening processing on the noise-reduced image data based on the edge sharpening parameter range to mark key edge feature points; Perform non-linear edge enhancement optimization on the key edge feature points to generate edge-enhanced image data.

4. The industrial vision recognition method based on deep learning according to claim 1, characterized in that, The specific steps of Step S2 are as follows: Step S21: Perform multi-scale convolutional network processing on the enhanced core wire welding image sequence to generate feature maps of different scales; Step S22: Perform solder joint feature recognition analysis on the feature maps of different scales to obtain multi-level solder joint feature representations; Step S23: Perform adaptive feature channel weighting based on the multi-level solder joint feature representations to generate a multi-level welding feature map; Step S24: Calculate the feature similarity of the multi-level welding feature map to obtain a feature similarity matrix; Step S25: Standardize the feature level differences of the multi-level welding feature map to generate a standardized multi-level feature map; Step S26: Perform cross-scale feature fusion on the standardized multi-level feature map based on the feature similarity matrix to generate a fused welding feature tensor.

5. The industrial vision recognition method based on deep learning according to claim 1, characterized in that, The specific steps of Step S3 are as follows: Step S31: Enhance the fused welding feature tensor through an attention mechanism to obtain a solder joint attention region tensor; Step S32: Automatically segment the solder joint region of the solder joint attention region tensor to obtain a solder joint region feature matrix; Step S33: Perform high-frequency detail enhancement processing on the solder joint region feature matrix to obtain a micro-detail enhanced feature matrix; Step S34: Perform micro-defect enhanced recognition on the micro-detail enhanced feature matrix to obtain a solder joint defect feature space.

6. The industrial vision recognition method based on deep learning according to claim 1, characterized in that The specific steps of Step S4 are as follows: Step S41: Obtain historical welding quality evaluation data and corresponding image annotation data; Step S42: Analyze the distribution of defect types in the historical welding quality evaluation data to generate a defect type distribution map; Step S43: Perform dynamic feature correlation analysis on the defect type distribution map to generate welding quality evaluation features; Step S44: Train a multi-level neural network on the welding quality evaluation features to construct a welding quality evaluation model.

7. The industrial vision recognition method based on deep learning according to claim 6, wherein The specific steps of Step S43 are as follows: Extract various defect feature parameters based on the defect type distribution map; Calculate the defect severity of the historical welding quality evaluation data according to various defect feature parameters to obtain a defect severity index; Conduct a defect spatial distribution analysis on the defect type distribution map to obtain defect spatial distribution characteristics; Perform a correlation analysis on the defect spatial distribution characteristics according to the defect severity index to obtain defect correlation feature data; Identify the influencing factors of welding process parameters based on the defect type distribution map; Conduct a defect cause correlation analysis based on the influencing factors of the process parameters to obtain process parameter correlation data; Perform dynamic feature fusion analysis on the process parameter correlation data and the defect correlation feature data to generate welding quality evaluation features.

8. The industrial vision recognition method based on deep learning according to claim 1, characterized in that The specific steps of Step S5 are as follows: Step S51: Use the welding quality evaluation model to perform multi-class defect recognition on the solder joint defect feature space and extract potential defect feature data; Step S52: Conduct a defect morphology analysis on the solder joint defect feature space to obtain a defect morphology feature link; Step S53: Perform multi-dimensional defect type recognition on the potential defect feature data based on the defect morphology feature link to obtain a defect type recognition result; Step S54: Quantify the quality impact degree of the defect type recognition result to generate a welding quality evaluation value; Step S55: Conduct an adaptive quality level evaluation based on the welding quality evaluation value to construct a welding quality diagnosis strategy.

9. The industrial vision recognition method based on deep learning according to claim 1, wherein The specific steps of Step S6 are as follows: Step S61: Construct a quality prediction model for the solder joint defect feature space to generate a welding quality prediction framework; Step S62: Map the diagnosis rules to the welding quality prediction framework based on the welding quality diagnosis strategy to construct a core wire welding quality diagnosis framework; Step S63: Perform dynamic quality prediction on the core wire welding quality diagnosis framework to construct a core wire welding quality real-time monitoring model.

10. An industrial vision recognition system based on deep learning, which is used to implement the industrial vision recognition method based on deep learning according to any one of claims 1 to 9, characterized in that, Including: A processing module, configured to obtain a multi-angle image dataset of core wire welding; Perform multi-scale noise reduction optimization on the multi-angle image dataset of core wire welding, and perform solder joint feature enhancement processing to obtain an enhanced core wire welding image sequence; A depth hierarchy module, configured to perform adaptive feature extraction on the enhanced core wire welding image sequence to generate a multi-level welding feature map; perform cross-scale feature fusion on the multi-level welding feature map to generate a fused welding feature tensor; A region space module, configured to automatically segment the solder joint region of the fused welding feature tensor to obtain a solder joint region feature matrix; perform microscopic defect enhancement recognition on the solder joint region feature matrix to obtain a solder joint defect feature space; A model construction module, configured to obtain historical welding quality assessment data; perform dynamic feature correlation analysis on the historical welding quality assessment data to construct a welding quality assessment model; A strategy fitting module, configured to use the welding quality assessment model to perform multi-dimensional defect type recognition on the solder joint defect feature space, and perform adaptive quality level assessment to construct a welding quality diagnosis strategy; A comprehensive module, based on the welding quality diagnosis strategy, performs dynamic quality prediction on the solder joint defect feature space to construct a real-time monitoring model for core wire welding quality; each module is connected by wired and / or wireless means.

Images