An industrial product quality inspection method and system based on machine vision

Abstract

The present disclosure relates to an industrial product quality inspection method and system based on machine vision, belonging to the technical field of quality inspection. During the operation of the system, through an industrial camera and an optimized light source layout, high-resolution images of electronic components are collected in real time by taking pictures from multiple angles. A series of optimization processes are performed on the collected images. According to the geometric characteristics of the electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination. Using the deep learning model algorithm, potential defects of the electronic components are identified, and after calculation, a quality evaluation index Q is obtained. According to the calculation results of feature extraction and defect detection, a comprehensive evaluation of the quality of each component is made. By comparing the quality evaluation index Q with the first preset threshold M and the second preset threshold N, a quality evaluation level is obtained, and the unqualified detection information is fed back to the production line to remove unqualified components.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection, and particularly to an industrial product quality inspection method and system based on machine vision. Background Art

[0002] Electronic components are widely used in modern industrial products, and the stability and reliability of their quality directly affect the overall performance of the products. However, due to the small size and complex structure of electronic components, combined with the precision requirements in their production process, traditional quality inspection methods often cannot meet the high-precision and high-efficiency inspection requirements. In this context, machine vision technology has gradually attracted attention. By using image acquisition and analysis technologies to achieve automatic recognition of product appearance features, it provides high-resolution and multi-angle real-time quality inspection means for electronic components. The application of machine vision technology has significantly improved the precision, efficiency, and consistency of the inspection process, and has gradually become an important means in industrial product quality inspection.

[0003] Traditional inspection methods mostly rely on manual labor or are completed through relatively inefficient mechanical means, which are not only low in efficiency but also poor in accuracy, and cannot ensure a high degree of consistency in product quality. During real-time inspection, the system can identify potential defects such as poor solder joints, missing components, and misaligned pins, effectively avoiding omissions and misjudgments in traditional inspections, and improving the yield and stability of the production line. The disadvantages of traditional quality inspection methods mainly stem from the deficiencies in inspection precision and efficiency, as well as the lack of real-time monitoring, resulting in difficulty in detecting some subtle defects and even being unable to achieve closed-loop feedback control of product quality. When abnormal situations or defect accumulations occur during the production process, they often cannot be identified and measures taken in a timely manner, which may ultimately lead to problems such as poor product consistency and high rework rates. This not only has a negative impact on production efficiency but also increases production costs and affects the competitiveness of enterprises in the market. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides an industrial product quality inspection method and system based on machine vision, which solves the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: An industrial product quality inspection system based on machine vision includes an image acquisition module, an image preprocessing module, a feature extraction module, a defect detection module, a quality evaluation module, and a control feedback module;

[0006] The image acquisition module is used to perform high-resolution image acquisition on electronic components in real time by using an industrial camera and an optimized light source layout, and taking multi-angle shots. The arrangement of the light source takes into account the small size and high-precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequencies, and subtle features of solder joints;

[0007] The image preprocessing module is used to perform a series of optimization processes on the collected images, including filtering for noise reduction, edge enhancement, and grayscale conversion. It uses background difference technology to eliminate the interference of ambient light and improve the contrast of the images.

[0008] The feature extraction module is used to extract the key features of electronic components, including the shape of solder joints, the arrangement of pins, and the position of components. According to the geometric characteristics of electronic components, it extracts shape, texture, and color features to form key parameters for defect discrimination.

[0009] The defect detection module is used to identify potential defects of electronic components using deep learning model algorithms, including poor solder joints, missing components, misaligned pins, and scratches. After calculation, it obtains: the quality evaluation index Q.

[0010] The quality evaluation module is used to make a comprehensive evaluation of the quality of each component based on the calculation results of feature extraction and defect detection. By comparing the quality evaluation index Q with the first preset threshold M and the second preset threshold N, it evaluates the size deviation, defect type, and quantity to obtain the quality evaluation grade.

[0011] The control feedback module is used to feedback the unqualified detection information to the production line, remove unqualified components, and adjust production parameters to improve the yield. By docking with the production management system, it forms a closed-loop feedback to continuously optimize the production process.

[0012] Preferably, the image acquisition module includes an image acquisition unit, a defect type unit, and a feature parameter synchronous storage unit.

[0013] The image acquisition unit is used to acquire clear and low-noise component images through an industrial camera and an optimized light source setting, and obtain: the total number of pixels N in the image, the gray value I(i) of the i-th pixel, and the gray value I(i + 1) of adjacent pixels.

[0014] The defect type unit is used to extract the defect frequency and features based on production history and real-time image data, use a known sample library to mark the occurrence frequency of different defects, record the distribution of various defects in the production process, and extract the original data of key feature parameters, including solder joints, pin positions, and arrangements, to obtain: the number of defect types M.

[0015] The feature parameter synchronous storage unit is used to synchronously store the image data and feature parameters, generate a unique identifier for each component, classify and file the data, and store the image data, defect marks, and feature data separately.

[0016] Preferably, the image preprocessing module includes an image preprocessing unit.

[0017] The image preprocessing unit is used to denoise the acquired image, reduce the interference noise in the image, ensure the image quality, enhance the edges in the image, make the boundaries of components and solder joints clearer for feature extraction, adopt background difference technology to eliminate the influence of ambient light, improve the image contrast, and make the component features more obvious.

[0018] Preferably, the feature extraction module includes a geometric feature extraction unit, a texture feature extraction unit, and a color feature extraction unit;

[0019] The geometric feature extraction unit is used to extract the geometric features of components. The geometric features include the shape of solder joints and the arrangement of pins, and provide feature parameters according to the extracted geometric features;

[0020] The texture feature extraction unit is used to analyze the texture features in the image, including the texture differences between solder joints and component surfaces, and provide support for defect detection;

[0021] The color feature extraction unit is used to extract color distribution features and record the color information of components for judging surface contamination or other abnormal conditions.

[0022] Preferably, the defect detection module includes a deep learning model unit and a quality evaluation index calculation unit;

[0023] The deep learning model unit is used to apply the trained deep learning model to automatically identify various defects in the image, classify the identified defects, including defective solder joints, missing components, misaligned pins, and surface scratches;

[0024] The quality evaluation index calculation unit is used to train a convolutional neural network (CNN) or a deep residual network (ResNet) model to automatically identify surface scratches, pits, and cracks, and calculate to obtain: the image clarity coefficient C, the defect probability coefficient P defect and the quality evaluation index Q of the component.

[0025] Preferably, the image clarity coefficient C is calculated and obtained through the following formula:

[0026] ;

[0027] In the formula, C represents the image clarity coefficient, N represents the total number of pixels in the image, I(i) represents the gray value of the i-th pixel, and I(i + 1) represents the gray value of the adjacent pixel;

[0028] The defect probability coefficient P defect is calculated and obtained through the following formula:

[0029] ;

[0030] In the formula, Pdefect represents the defect probability coefficient, M represents the number of defect types, and w j represents the weight of the j-th type of defect, and F j represents the occurrence frequency of the j-th type of defect.

[0031] Preferably, the quality evaluation index Q of the component is obtained by calculating the following formula:

[0032] ;

[0033] In the formula, Q represents the quality evaluation index of the component, C represents the image clarity coefficient, P defect represents the defect probability coefficient, and T k represents the quality evaluation value of the characteristic parameter k, K represents the number of key characteristics of the quality inspection, represents the weight coefficient.

[0034] Preferably, the quality evaluation module includes an evaluation level unit;

[0035] The evaluation level unit is used to compare the calculated quality evaluation index Q with the preset first threshold M and second threshold A, determine whether the component is qualified, evaluate the dimensional deviation, defect type and quantity of the component, generate quality analysis data, and classify the quality of the component based on the comprehensive evaluation result, providing a specific basis for production quality control;

[0036] If the quality evaluation index Q ≥ the first threshold M, obtain the first evaluation level, indicating that the quality of the component meets the qualified standard, the solder joint shape, pin arrangement, and color uniformity meet the requirements, no further operation is required, mark this batch of components as qualified and record them in the system database;

[0037] If the second threshold A ≤ the quality evaluation index Q < the first threshold M, obtain the second evaluation level. This interval is the warning level, indicating that there are defect problems within 25% of the component, manifested as slight solder joint defects or uneven color distribution, but it does not affect the overall function. Mark it as the warning level, divide this batch of components into an independent area and trigger the maintenance program, and the quality inspection system issues a warning and feedbacks it to the production line for adjustment;

[0038] If the quality evaluation index Q < the second threshold A, obtain the third evaluation level. Such components have serious defects and cannot meet the functional requirements. Immediately remove the unqualified components, record the specific defect types and feedback the data to the production management system for defect analysis. Further adjust the production equipment and process parameters through the control feedback module. For the identified defect types and characteristic parameters, formulate defect prevention strategies. For surface scratches and cracks, increase the quality inspection frequency.

[0039] Preferably, the control feedback module includes a non - compliance feedback unit and a production parameter adjustment unit;

[0040] The non - compliance feedback unit is used to feed back the detected information of non - compliant components to the production line, promptly remove products that do not meet the quality standards, transmit non - compliance information in real - time, effectively prevent defective products from flowing into downstream processes, and ensure that the production line takes prompt measures when quality problems occur;

[0041] The production parameter adjustment unit is used to adjust the process parameters of the production line according to the defect type, frequency, and distribution data in the quality assessment results, including welding temperature, equipment pressure, or light source layout, ensure that the production process meets the product quality requirements, form a closed - loop feedback system, and gradually reduce the defect rate by adjusting production parameters.

[0042] An industrial product quality detection method based on machine vision includes the following steps:

[0043] Step 1: Through an industrial camera and an optimized light source layout, high - resolution images of electronic components are collected in real - time by taking multi - angle shots. The layout of the light source takes into account the small size and high - precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequency, and fine features of welding points;

[0044] Step 2: A series of optimization processes are performed on the collected images, including filtering and noise reduction, edge enhancement, and grayscale conversion. The background difference technology is used to eliminate the interference of ambient light and improve the contrast of the images;

[0045] Step 3: Key features of electronic components are extracted, including the shape of welding points, the arrangement of pins, and the position of components. According to the geometric characteristics of electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination;

[0046] Step 4: Using deep - learning model algorithms, potential defects of electronic components are identified, including poor solder joints, missing components, misaligned pins, and scratches. After calculation, the quality assessment index Q is obtained;

[0047] Step 5: Based on the calculation results of feature extraction and defect detection, a comprehensive evaluation of the quality of each component is made. By comparing the quality assessment index Q with the first preset threshold M and the second preset threshold N, size deviation, defect type, and quantity are evaluated to obtain the quality assessment level;

[0048] Step 6: Feed back the non - compliant detection information to the production line, remove non - compliant components, and adjust production parameters to improve the yield rate. By docking with the production management system, a closed - loop feedback is formed to continuously optimize the production process.

[0049] The present invention provides an industrial product quality inspection method and system based on machine vision, which has the following beneficial effects:

[0050] 1) When the system runs, through an industrial camera and an optimized light source layout, high-resolution images of electronic components are collected in real time by taking pictures from multiple angles. A series of optimization processes are performed on the collected images. According to the geometric characteristics of the electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination. Using the deep learning model algorithm, potential defects of the electronic components are identified, and after calculation, the quality evaluation index Q is obtained. According to the calculation results of feature extraction and defect detection, a comprehensive evaluation of the quality of each component is made. By comparing the quality evaluation index Q with the first preset threshold M and the second preset threshold N, the quality evaluation level is obtained, and the unqualified detection information is fed back to the production line to remove unqualified components.

[0051] 2) Through the coordinated operation of modules such as image acquisition, image preprocessing, feature extraction, defect detection, quality evaluation, and control feedback, high-precision quality inspection of electronic components in all directions and multiple dimensions is realized. The image acquisition module and the image preprocessing module enable the system to obtain clear and high-resolution image data and eliminate environmental interference to highlight the detailed features of the components. The feature extraction module further analyzes and records the geometric shape, texture, and color features of the product as important parameter bases for defect identification. These steps lay a solid foundation for subsequent defect detection and quality evaluation.

[0052] 3) By introducing the deep learning model, the defect detection module of this system can quickly and accurately identify various defect types and further generate the quality evaluation index Q. This index combines parameters such as the image clarity coefficient and the defect probability coefficient to quantitatively evaluate the quality of the components, thereby realizing the comprehensive evaluation ability of the quality evaluation module. On this basis, the system outputs quality levels hierarchically according to different evaluation results to support precise production quality control. Compared with traditional detection means, this evaluation mechanism not only improves the accuracy of defect identification and classification precision through the application of deep learning technology, but also realizes the automation and real-time of the detection process, enabling the production line to adjust the process in a timely manner according to the evaluation data.

[0053] (4) The system feeds back the detection information to the production line in real time through the control feedback module, realizes the rejection of unqualified products and the optimization and adjustment of production parameters, and forms a closed-loop feedback control system. Compared with traditional means, this system significantly reduces the error rate and missed inspection rate of human intervention, while improving the overall production quality and consistency. By reducing the defective rate and rework cost, this system effectively improves the production efficiency and finished product rate, enhances the reliability and automation degree of the production process, and achieves the goals of product quality improvement and production efficiency optimization. This intelligent quality inspection system not only improves the limitations of traditional inspection methods, but also promotes the quality control level of high-precision products such as electronic components in industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a block diagram of an industrial product quality inspection system based on machine vision of the present invention;

[0055] Figure 2 It is a schematic diagram of the steps of an industrial product quality inspection method based on machine vision of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] 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.

[0057] Embodiment 1

[0058] The present invention provides an industrial product quality inspection system based on machine vision. Please refer to Figure 1 , including an image acquisition module, an image preprocessing module, a feature extraction module, a defect detection module, a quality evaluation module and a control feedback module;

[0059] The image acquisition module is used to collect high-resolution images of electronic components in real time by taking multi-angle shots through an industrial camera and an optimized light source layout. The layout of the light source takes into account the small size and high-precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequencies and fine features of welding points;

[0060] The image preprocessing module is used to perform a series of optimization processes on the collected images, including filtering and noise reduction, edge enhancement and grayscale conversion, and uses background difference technology to eliminate the interference of ambient light and improve the contrast of the images;

[0061] The feature extraction module is used to extract the key features of electronic components, including the shape of solder joints, the arrangement of pins, and the position of components. According to the geometric characteristics of electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination.

[0062] The defect detection module is used to identify potential defects of electronic components, including poor solder joints, missing components, misaligned pins, and scratches, using deep learning model algorithms. After calculation, the quality evaluation index Q is obtained.

[0063] The quality evaluation module is used to make a comprehensive evaluation of the quality of each component based on the calculation results of feature extraction and defect detection. By comparing the quality evaluation index Q with the first preset threshold M and the second preset threshold N, size deviation, defect type, and quantity are evaluated to obtain the quality evaluation grade.

[0064] The control feedback module is used to feedback unqualified detection information to the production line, remove unqualified components, and adjust production parameters to improve the yield. By docking with the production management system, a closed-loop feedback is formed to continuously optimize the production process.

[0065] In this embodiment, through an industrial camera and an optimized light source layout, high-resolution images of electronic components are collected in real time from multiple angles. The arrangement of the light source takes into account the small size and high-precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequency, and fine features of solder joints. A series of optimization processes are performed on the collected images, including filtering and noise reduction, edge enhancement, and grayscale conversion. The background difference technology is used to eliminate the interference of ambient light, improve the contrast of the image, extract the key features of electronic components, including the shape of solder joints, the arrangement of pins, and the position of components. According to the geometric characteristics of electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination. Using deep learning model algorithms, potential defects of electronic components, including poor solder joints, missing components, misaligned pins, and scratches, are identified. After calculation, the quality evaluation index Q is obtained. Based on the calculation results of feature extraction and defect detection, a comprehensive evaluation of the quality of each component is made. By comparing the quality evaluation index Q with the first preset threshold M and the second preset threshold N, size deviation, defect type, and quantity are evaluated to obtain the quality evaluation grade. Unqualified detection information is feedback to the production line, unqualified components are removed, and production parameters are adjusted to improve the yield. By docking with the production management system, a closed-loop feedback is formed to continuously optimize the production process.

[0066] Embodiment 2

[0067] This embodiment is an explanatory description based on Embodiment 1. Please refer to Figure 1, specifically: the image acquisition module includes an image acquisition unit, a defect type unit, and a feature parameter synchronous storage unit;

[0068] The image acquisition unit is used to collect clear and low-noise component images through an industrial camera and an optimized light source setting, and obtain: the total number of pixels N in the image, the gray value I(i) of the i-th pixel, and the gray value I(i + 1) of adjacent pixels;

[0069] The defect type unit is used to extract defect frequencies and characteristics based on production history and real-time image data, use a known sample library to mark the occurrence frequencies of different defects, record the distribution of various defects in the production process, and extract the original data of key feature parameters, including solder joints, pin positions, and arrangements, and obtain: the number of defect types M;

[0070] The feature parameter synchronous storage unit is used to synchronously store image data and feature parameters, generate a unique identifier for each component, classify and file the data, and store the image data, defect marks, and feature data separately.

[0071] The image preprocessing module includes an image preprocessing unit;

[0072] The image preprocessing unit is used to perform denoising processing on the collected image, reduce the interference noise in the image, ensure the image quality, enhance the edges in the image, make the boundaries of the components and the solder joints clearer for feature extraction, adopt background difference technology to eliminate the influence of ambient light, improve the image contrast, and make the component features more obvious.

[0073] The feature extraction module includes a geometric feature extraction unit, a texture feature extraction unit, and a color feature extraction unit;

[0074] The geometric feature extraction unit is used to extract the geometric features of the components. The geometric features include the shapes of the solder joints and the pin arrangements, and provide feature parameters according to the extracted geometric features;

[0075] The texture feature extraction unit is used to analyze the texture features in the image, including the texture differences between the solder joints and the component surfaces, and provide support for defect detection;

[0076] The color feature extraction unit is used to extract color distribution features and record the color information of the components for judging surface contamination or other abnormal conditions.

[0077] In this embodiment, through the close cooperation of the image acquisition, preprocessing, and feature extraction modules, the accuracy and efficiency of the quality inspection of electronic components are significantly improved. The image acquisition unit in the image acquisition module, combined with the optimized light source settings, ensures that the acquired component images are clear and have low noise, which helps to highlight the tiny features of the product. The introduction of the defect type unit enables the system to extract the defect types and distribution frequencies based on the production history and real-time image data. Through the management of the feature parameter synchronization storage unit, the system can generate a unique identifier for each component, realizing the effective classification and archiving of data. The image preprocessing module further improves the image quality. Through denoising and edge enhancement, the key details such as boundaries and solder joints in the image become clearer. In particular, the application of the background difference technology effectively eliminates the interference of ambient light, making the features of the components more distinct, thereby improving the reliability of subsequent inspections. The geometric feature, texture feature, and color feature extraction units in the feature extraction module provide multi-dimensional feature support for defect detection. They can not only extract the geometric and surface information of the components but also monitor color abnormalities to quickly judge surface contamination or other potential problems.

[0078] Embodiment 3

[0079] This embodiment is an explanatory description based on Embodiment 1. Please refer to Figure 1 , specifically: The defect detection module includes a deep learning model unit and a quality evaluation index calculation unit;

[0080] The deep learning model unit is used to apply the trained deep learning model to automatically identify various defects in the image and classify the identified defects, including poor solder joints, missing components, pin misalignment, and surface scratches;

[0081] The quality evaluation index calculation unit is used to train a convolutional neural network (CNN) or a deep residual network (ResNet) model to automatically identify surface scratches, pits, and cracks, and calculate to obtain: the image clarity coefficient C, the defect probability coefficient P defect and the quality evaluation index Q of the component.

[0082] The image clarity coefficient C is calculated and obtained through the following formula:

[0083] ;

[0084] In the formula, C represents the image clarity coefficient, N represents the total number of pixels in the image, I(i) represents the gray value of the i-th pixel, and I(i + 1) represents the gray value of the adjacent pixel;

[0085] The defect probability coefficient P defect is calculated and obtained through the following formula:

[0086] ;

[0087] Wherein, P defect represents the defect probability coefficient, M represents the number of defect types, and w j represents the weight of the j-th type of defect, and F j represents the occurrence frequency of the j-th type of defect.

[0088] The quality evaluation index Q of the component is obtained by calculating the following formula:

[0089] ;

[0090] Wherein, Q represents the quality evaluation index of the component, C represents the image clarity coefficient, P defect represents the defect probability coefficient, T k represents the quality evaluation value of the characteristic parameter k, K represents the number of key characteristics of the quality inspection, represents the weight coefficient.

[0091] In this embodiment, the defect detection module of the system realizes the automatic and high-precision identification of electronic component defects through the combination of a deep learning model unit and a quality evaluation index calculation unit. The deep learning model unit uses the trained model to quickly classify various types of defects in the image, covering key defect types such as poor solder joints, missing components, misaligned pins, and surface scratches. The quality evaluation index calculation unit further accurately identifies more subtle defects such as surface scratches, pits, and cracks through a convolutional neural network (CNN) or a deep residual network (ResNet), generating key indicators such as the clarity coefficient C and the defect probability coefficient P defect to provide a quantitative basis for comprehensive quality evaluation.

[0092] Embodiment 4

[0093] This embodiment is an explanatory description carried out in Embodiment 1. Please refer to Figure 1 , specifically: The quality evaluation module includes an evaluation level unit;

[0094] The evaluation level unit is used to compare the calculated quality evaluation index Q with a preset first threshold M and a second threshold A, determine whether the component is qualified, evaluate the dimensional deviation, defect type and quantity of the component, generate quality analysis data, and classify the quality of the component based on the result of the comprehensive evaluation, providing a specific basis for production quality control;

[0095] If the quality evaluation index Q ≥ the first threshold M, obtain the first evaluation level, indicating that the quality of the component meets the qualified standard, the solder joint shape, pin arrangement, and color uniformity meet the requirements, no further operation is required, mark this batch of components as qualified and record them in the system database;

[0096] If the second threshold A ≤ quality assessment index Q < the first threshold M, obtain the second assessment level. This range is the warning level, indicating that there are defect problems within 25% in the components, manifested as minor solder joint defects or uneven color distribution, but it does not affect the overall function yet. Mark it as the warning level, divide the components of this batch into independent areas and trigger the maintenance procedure. The quality inspection system issues a warning and feeds it back to the production line for adjustment;

[0097] If the quality assessment index Q < the second threshold A, obtain the third assessment level. Such components have serious defects and cannot meet the functional requirements. Immediately reject the unqualified components, record the specific defect types and feed the data back to the production management system for defect analysis. Further adjust the production equipment and process parameters through the control feedback module. For the identified defect types and characteristic parameters, formulate defect prevention strategies. For surface scratch and crack problems, increase the quality inspection frequency.

[0098] The control feedback module includes an unqualified feedback unit and a production parameter adjustment unit;

[0099] The unqualified feedback unit is used to feed back the information of the detected unqualified components to the production line, promptly reject the products that do not meet the quality standards, transmit the unqualified information in real time, effectively prevent defective products from flowing into the downstream links, and ensure that the production line takes measures promptly when quality problems occur;

[0100] The production parameter adjustment unit is used to adjust the process parameters of the production line according to the defect types, frequencies and distribution data in the quality assessment results, including welding temperature, equipment pressure or light source layout, ensure that the production process meets the product quality requirements, form a closed-loop feedback system, and gradually reduce the defect rate by adjusting the production parameters.

[0101] In this embodiment, the quality assessment module and the control feedback module of the system effectively improve the accuracy and response speed of product quality management through a strict grading and feedback mechanism. The quality assessment module compares the quality assessment index Q of each component with the preset thresholds M and A through the assessment grading unit, accurately divides the quality grades, and thus provides a clear operation basis for production control. For components of the qualified grade, the system can automatically file the data; while components in the warning grade trigger the maintenance procedure to ensure that minor defects are repaired in time; unqualified components are immediately removed to prevent defective products from flowing into the subsequent links and ensure product quality consistency. The unqualified feedback unit in the control feedback module enables the production line to quickly respond to quality problems by transmitting unqualified information in real time, greatly reducing the circulation of defective products. The production parameter adjustment unit dynamically optimizes production parameters such as welding temperature and equipment pressure based on defect data, forming a closed-loop feedback system, effectively reducing the defect rate. Compared with traditional quality control methods, the system can significantly improve the adjustment speed and optimization ability of the production line through quantitative data and feedback mechanism, reduce the rework rate, further improve production efficiency and product yield, and achieve the beneficial effects of intelligent quality control and continuous optimization.

[0102] Embodiment 5

[0103] An industrial product quality inspection method based on machine vision, please refer to Figure 2 , specifically: including the following steps:

[0104] Step 1: Through an industrial camera and an optimized light source layout, high-resolution images of electronic components are collected in real time from multiple angles. The arrangement of the light source takes into account the small size and high-precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequency, and fine features of welding points;

[0105] Step 2: A series of optimization processes are performed on the collected images, including filtering and noise reduction, edge enhancement, and grayscale conversion. The background difference technology is used to eliminate the interference of ambient light and improve the contrast of the images;

[0106] Step 3: Key features of the electronic components are extracted, including the shape of the welding points, the arrangement of the pins, and the position of the components. According to the geometric characteristics of the electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination;

[0107] Step 4: Using the deep learning model algorithm, potential defects of the electronic components are identified, including poor welding points, missing components, misaligned pins, and scratches. After calculation, the quality assessment index Q is obtained;

[0108] Step 5: Based on the calculation results of feature extraction and defect detection, make a comprehensive evaluation of the quality of each component. By comparing the quality evaluation index Q with the first preset threshold M and the second preset threshold N, evaluate the dimensional deviation, defect type and quantity, and obtain the quality evaluation level.

[0109] Step 6: Feed back the unqualified detection information to the production line, remove the unqualified components, and adjust the production parameters to improve the yield. By docking with the production management system, form a closed-loop feedback to continuously optimize the production process.

[0110] In this embodiment, the system realizes efficient and accurate quality control of electronic components through six systematic steps, significantly improving the reliability of detection and the yield of the production line. First, through multi-angle image acquisition using industrial cameras and optimized light source layouts, the subtle features in the images are fully presented. In the subsequent image preprocessing step, environmental interference is eliminated through denoising and edge enhancement to ensure that the image quality is more suitable for analysis and processing. The key feature extraction step lays a solid foundation for defect recognition from aspects such as shape, texture, and color. The introduction of the deep learning model enables automatic recognition during the defect detection process, quickly and accurately classifying defect types such as poor solder joints and missing components, and quantifying them as the quality evaluation index Q. In the comprehensive evaluation stage, the system compares Q with the preset threshold to determine the quality level of each component, thus more efficiently determining the qualification of the product. Finally, by feeding back the detection information to the production line, removing the unqualified components and optimizing the production parameters, a closed-loop feedback process is formed. Compared with traditional methods, the systematic detection process not only improves the detection accuracy and efficiency, but also realizes continuous production improvement through automatic feedback and real-time optimization, significantly enhancing the product consistency and the intelligent level of the production process.

[0111] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art 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 appended claims and their equivalents.

Claims

1. An industrial product quality inspection system based on machine vision, characterized in that: It includes an image acquisition module, an image preprocessing module, a feature extraction module, a defect detection module, a quality assessment module, and a control feedback module; The image acquisition module is used to perform high-resolution image acquisition on electronic components in real time by taking multi-angle shots through an industrial camera and an optimized light source layout. The layout of the light source takes into account the small size and high-precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequency, and fine features of solder joints; The image preprocessing module is used to perform a series of optimization processes on the acquired images, including filtering and noise reduction, edge enhancement, and grayscale conversion. The background difference technology is used to eliminate the interference of ambient light and improve the contrast of the images; The feature extraction module is used to extract the key features of electronic components, including the shape of solder joints, the arrangement of pins, and the position of components. According to the geometric characteristics of electronic components, shape, texture, and color features are extracted to form key parameters for defect discrimination; The defect detection module is used to use deep learning model algorithms to identify potential defects of electronic components, including poor solder joints, missing components, pin misalignment, and scratches. After calculation, the quality assessment index Q is obtained; The quality assessment module is used to make a comprehensive evaluation of the quality of each component according to the calculation results of feature extraction and defect detection. By comparing the quality assessment index Q with the first preset threshold M and the second preset threshold N, size deviation, defect type, and quantity are evaluated to obtain the quality assessment level; The control feedback module is used to feedback unqualified detection information to the production line, remove unqualified components, and adjust production parameters to improve the yield. By docking with the production management system, a closed-loop feedback is formed to continuously optimize the production process; The image clarity coefficient C is obtained by calculating through the following formula: ； In the formula, C represents the image clarity coefficient, N represents the total number of pixels in the image, I(i) represents the gray value of the i-th pixel, and I(i + 1) represents the gray value of the adjacent pixel; Defect probability coefficient P defect Obtained by calculation using the following formula: ； Wherein, P defect represents the defect probability coefficient, M represents the number of defect types, w j represents the weight of the j-th type of defect, and F j represents the occurrence frequency of the j-th type of defect; The quality assessment index Q of the component is obtained by calculating through the following formula: ； Wherein, Q represents the quality evaluation index of the component, C represents the image clarity coefficient, and P defect represents the defect probability coefficient, T k represents the quality evaluation value of the characteristic parameter k, K represents the number of key features for quality inspection, represents the weight coefficient.

2. The industrial product quality inspection system based on machine vision according to claim 1, characterized in that: The image acquisition module includes an image acquisition unit, a defect type unit, and a feature parameter synchronous storage unit; The image acquisition unit is used to acquire clear and low-noise component images through an industrial camera and an optimized light source setting, and obtain: the total number of pixels N in the image, the gray value I(i) of the i-th pixel, and the gray value I(i + 1) of the adjacent pixel; The defect type unit is used to extract defect frequency and features based on production history and real-time image data, use a known sample library to mark the occurrence frequency of different defects, record the distribution of various defects in the production process, and extract the original data of key feature parameters, including solder joints, pin positions, and arrangements, to obtain: the number of defect types M; The feature parameter synchronous storage unit is used to synchronously store image data and feature parameters, generate a unique identifier for each component, classify and file the data, and store the image data, defect marks, and feature data separately.

3. An industrial product quality inspection system based on machine vision according to claim 1, characterized in that: The image preprocessing module includes an image preprocessing unit; The image preprocessing unit is used to denoise the collected images, reduce the interference noise in the images, ensure the image quality, enhance the edges in the images, make the boundaries of components and solder joints clearer for feature extraction, adopt background difference technology to eliminate the influence of ambient light, improve the image contrast, and make the component features more obvious.

4. An industrial product quality inspection system based on machine vision according to claim 1, characterized in that: The feature extraction module includes a geometric feature extraction unit, a texture feature extraction unit, and a color feature extraction unit; The geometric feature extraction unit is used to extract the geometric features of components. The geometric features include the shape of solder joints and the arrangement of pins, and provide feature parameters according to the extracted geometric features; The texture feature extraction unit is used to analyze the texture features in the images, including the texture differences between solder joints and component surfaces, and provide support for defect detection; The color feature extraction unit is used to extract the color distribution features and record the color information of components for judging surface contamination or other abnormal conditions.

5. An industrial product quality inspection system based on machine vision according to claim 1, characterized in that: The defect detection module includes a deep learning model unit and a quality assessment index calculation unit; The deep learning model unit is used to apply the trained deep learning model to automatically identify various defects in the images, classify the identified defects, including poor solder joints, missing components, misaligned pins, and surface scratches; The quality assessment index calculation unit is used to train a convolutional neural network (CNN) or a deep residual network (ResNet) model to automatically identify surface scratches, pits, and cracks, and after calculation, obtain: the image clarity coefficient C, the defect probability coefficient P defect and the quality assessment index Q of the component.

6. An industrial product quality inspection system based on machine vision according to claim 1, characterized in that: The quality assessment module includes an assessment level unit; The assessment level unit is used to compare the calculated quality assessment index Q with the preset first threshold M and second threshold A to determine whether the components are qualified, evaluate the dimensional deviation, defect type, and quantity of the components, generate quality analysis data, and classify the quality of the components based on the results of the comprehensive assessment, providing specific basis for production quality control; If the quality assessment index Q ≥ the first threshold M, obtain the first assessment level, indicating that the quality of the components meets the qualified standard, the solder joint shape, pin arrangement, and color uniformity meet the requirements, no further operation is required, mark this batch of components as qualified and record them in the system database; If the second threshold A ≤ the quality assessment index Q < the first threshold M, obtain the second assessment level. This interval is the warning level, indicating that there are defect problems within 25% of the components, manifested as minor solder joint defects or uneven color distribution, but it does not affect the overall function yet. Mark it as the warning level, divide this batch of components into an independent area and trigger the maintenance program, and the quality inspection system issues a warning and feedbacks it to the production line for adjustment; If the quality assessment index Q < the second threshold A, obtain the third assessment level. Such components have serious defects and cannot meet the functional requirements. Immediately reject the unqualified components, record the specific defect types, and feedback the data to the production management system for defect analysis. Further adjust the production equipment and process parameters through the control feedback module. According to the identified defect types and feature parameters, formulate defect prevention strategies. For surface scratches and cracks, increase the quality inspection frequency.

7. An industrial product quality inspection system based on machine vision according to claim 1, characterized in that: The control feedback module includes an unqualified feedback unit and a production parameter adjustment unit; The unqualified feedback unit is used to feedback the detected unqualified component information to the production line, promptly remove the products that do not meet the quality standards, transmit the unqualified information in real time, effectively prevent defective products from flowing into the downstream links, and ensure that the production line takes prompt measures when quality problems occur; The production parameter adjustment unit is used to adjust the process parameters of the production line according to the defect type, frequency and distribution data in the quality assessment result, including welding temperature, equipment pressure or light source layout, ensure that the production process meets the product quality requirements, form a closed-loop feedback system, and gradually reduce the defect rate by adjusting the production parameters.

8. A method for detecting the quality of industrial products based on machine vision, which is applied to a system for detecting the quality of industrial products based on machine vision according to any one of claims 1 to 7, characterized in that: It includes the following steps: Step 1: Through an industrial camera and an optimized light source layout, high-resolution image acquisition of electronic components is carried out in real time by taking pictures from multiple angles. The layout of the light source takes into account the small size and high-precision requirements of electronic components, and can highlight image pixels, clarity, product edges, defect frequency and subtle features of welding points; Step 2: A series of optimization processes are performed on the acquired images, including filtering and noise reduction, edge enhancement and grayscale conversion. The background difference technology is used to eliminate the interference of ambient light and improve the contrast of the images; Step 3: Extract the key features of the electronic components, including the shape of the welding points, the arrangement of the pins and the position of the components. According to the geometric characteristics of the electronic components, extract the shape, texture and color features to form the key parameters for defect discrimination; Step 4: Use the deep learning model algorithm to identify the potential defects of the electronic components, including poor welding points, missing components, misaligned pins and scratches, and calculate to obtain the quality assessment index Q; Step 5: Based on the calculation results of feature extraction and defect detection, make a comprehensive evaluation of the quality of each component. By comparing the quality assessment index Q with the first preset threshold M and the second preset threshold N, evaluate the size deviation, defect type and quantity, and obtain the quality assessment grade; Step 6: Feedback the unqualified detection information to the production line, remove the unqualified components, and adjust the production parameters to improve the yield. By docking with the production management system, a closed-loop feedback is formed to continuously optimize the production process.

Images