High-frequency high-speed circuit board detection method based on infrared image enhancement and mode recognition

Abstract

The invention discloses a high-frequency and high-speed circuit board detection method based on infrared image enhancement and mode recognition, and relates to the technical field of circuit board defect detection, and the method comprises the steps: carrying out the non-contact infrared image collection of a high-frequency and high-speed circuit board which is electrified to work through a thermal infrared imager, and obtaining the original infrared image data containing the temperature distribution of each region of the circuit board; performing enhancement processing on the infrared image by applying histogram equalization and filtering denoising algorithms, and highlighting a tiny thermal change region; and extracting heat distribution characteristics from the enhanced infrared image to obtain a heat distribution characteristic sequence table. By combining infrared image enhancement and mode recognition and adopting a non-contact detection mode, physical damage to the circuit board caused by traditional contact probe detection is avoided, the integrity of the circuit board is protected, the service life of the circuit board is prolonged, the defective rate caused by detection is reduced, and the detection efficiency is improved. The method is particularly important for high-precision and high-value products such as high-frequency and high-speed circuit boards.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit board defect detection, and in particular to a high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition. Background Art

[0002] As the requirements for high performance, high frequency and high speed in fields such as communications, computers, automotive electronics, aerospace, etc. continue to increase, the design and manufacturing of circuit boards are facing increasingly severe challenges. With the promotion of 5G technology and the research on 6G in the future, high-frequency and high-speed circuit boards play a key role in wireless communication systems. Circuit boards need to support higher transmission rates and wider bandwidths, and require strict control and detection of signal quality, transmission delay, noise, etc. Therefore, the detection of high-frequency and high-speed circuit boards has become particularly important.

[0003] In the prior art, traditional electrical inspection of high-frequency, high-speed circuit boards often uses contact probe inspection. This method can cause physical damage to the circuit boards, and tiny defects in solder joints or other surface problems cannot be fully discovered, making it difficult to accurately locate or completely inspect all solder joints. Therefore, how to use a combination of infrared image enhancement and pattern recognition to perform non-contact inspection that can accurately display tiny thermal changes on the circuit board and provide high-precision defect diagnosis is the problem to be solved by the present invention. To this end, a high-frequency, high-speed circuit board inspection method based on infrared image enhancement and pattern recognition is proposed. Summary of the Invention

[0004] The present invention aims to provide a high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition includes the following steps:

[0007] S1. Use an infrared thermal imager to perform non-contact infrared image acquisition on a high-frequency, high-speed circuit board that is powered on, and obtain raw infrared image data containing the temperature distribution of each area of ​​the circuit board;

[0008] S2. Use histogram equalization and filtering denoising algorithms to enhance the infrared image, highlight the area of ​​small thermal changes, and improve image clarity and contrast;

[0009] S3, extracting thermal distribution features from the enhanced infrared image to obtain a thermal distribution feature sequence table as input data for subsequent pattern recognition;

[0010] S4. Collect a large number of infrared image samples of normal and defective circuit boards, mark the defect types and locations, and pre-train the defect recognition model based on the deep learning algorithm to enable it to have defect recognition capabilities;

[0011] S5. Input the thermal distribution features in the thermal distribution feature sequence table into the defect recognition model, analyze and classify the extracted features, identify the defect type, and combine the pattern recognition results to accurately locate the position of the defect on the circuit board, diagnose the defect based on the defect type, and evaluate the impact of the defect;

[0012] S6. Output a defect diagnosis report containing defect type and location information, and visualize the identified defect information.

[0013] A further improvement of the technical solution of the present invention is that: S1 specifically includes:

[0014] Choose a high-resolution, high-sensitivity infrared thermal imager. Check whether its lens, battery, storage device and other components are normal to ensure stable operation. Set up a relatively closed detection environment without strong interference light sources. At the same time, use a standard blackbody source to calibrate the infrared thermal imager to eliminate instrument errors and ensure that the measured temperature data is accurate and reliable.

[0015] Install the high-frequency and high-speed circuit board to be tested on the test fixture, ensuring that the circuit board is firmly installed and will not shake during the test process, which will affect the imaging quality. Provide it with power and working signals according to the design requirements of the circuit board to ensure that the circuit board is in normal working condition. After powering on, wait for a period of time for the circuit board to reach a stable working condition.

[0016] Based on the structure and layout of the circuit board, multiple shooting angles are planned to fully cover all areas of the circuit board to avoid missing important parts. The infrared thermal imager is used to shoot the high-frequency and high-speed circuit board that is powered on from the planned angles. While capturing the image, the image capture time, circuit board number, and shooting angle information are recorded. The collected raw infrared image data containing the temperature distribution of each area of ​​the circuit board is stored in the data warehouse to facilitate subsequent data processing and analysis.

[0017] After completing the image acquisition, check whether the original infrared image is complete and whether there is any missing or damaged information. If the image is found to be incomplete, re-acquire the image at that angle to ensure the integrity of the data, and then back up the original infrared image data that has passed the inspection. At the same time, organize and classify the data according to the circuit board number and shooting time, and establish a data directory.

[0018] A further improvement of the technical solution of the present invention is that: S2 specifically includes:

[0019] Import the collected original infrared image into the image processing software to check the overall image situation and preliminarily set the relevant parameters of the histogram equalization and filtering denoising algorithm according to the image characteristics;

[0020] The infrared image is processed using the selected histogram equalization algorithm. By stretching the grayscale range of the image, the details of the originally darker or brighter areas in the image are made clearer and the areas with slight thermal changes are highlighted. During the processing, the pixel grayscale values ​​are adjusted according to the grayscale histogram of the image to enhance the contrast of the image.

[0021] Perform filtering and denoising on the image after histogram equalization according to the initially set filtering algorithm and parameters. Check the filtered and denoised image to see if the noise is effectively removed and whether the edges and details of the image are overly blurred.

[0022] The infrared image after histogram equalization and filtering denoising is compared with the original infrared image, and a comprehensive evaluation is performed from multiple aspects such as clarity, contrast, and noise removal. The small thermal change areas are checked to see whether the temperature distribution characteristics of the circuit board are more obvious. Based on the comprehensive evaluation results, the parameters of the histogram equalization and filtering denoising algorithms are fine-tuned to further optimize the image processing effect. When a satisfactory enhancement effect is achieved, the processed infrared image is output and saved.

[0023] A further improvement of the technical solution of the present invention is that: S3 specifically includes:

[0024] Import the enhanced infrared image into the image analysis software and use the clustering algorithm to segment the image and distinguish different areas of the circuit board. Based on the circuit board layout and component distribution, further refine the segmentation results to divide the key areas of the chip, solder joints and traces. Each area is given a unique identifier, and the area number and location information are recorded to ensure that subsequent feature extraction can accurately correspond to the specific part. At the same time, the boundary information of each area is recorded.

[0025] For each divided area, basic thermal features including regional temperature mean, regional temperature standard deviation and temperature range are extracted, as well as advanced thermal features including temperature gradient, hot spot area and hot spot temperature peak;

[0026] Summarize the basic thermal characteristics and advanced thermal characteristics data of each area, integrate the extracted temperature mean, temperature standard deviation, temperature range, temperature gradient, hot spot area and hot spot temperature peak characteristics of each area into a thermal distribution feature sequence table, and sort and normalize the thermal distribution feature sequence table according to the order of the circuit board area to ensure that the feature sequence format is consistent and the information is complete.

[0027] A further improvement of the technical solution of the present invention is that: S4 specifically includes:

[0028] A large number of infrared image samples of high-frequency and high-speed circuit boards are collected from actual production environments, simulated test scenarios, and databases. These samples cover images of circuit boards of different models and working conditions. This ensures that the samples are diverse and fully represent various situations in actual applications, including normal circuit boards and those with various typical defects. The collected image samples are preliminarily screened and classified and stored according to normal and defective conditions.

[0029] Organize technical personnel to develop defect annotation specifications based on the characteristics of circuit board defects and actual needs. That is, defects are annotated according to the thermal distribution characteristics of the thermal distribution feature sequence table, and reference values ​​for each thermal distribution feature are set. Image annotation tools are used to annotate infrared image samples of defective circuit boards according to the annotation specifications. Defects in the images are identified, and the location and type of defects are marked. The annotated sample data set is divided into training set, validation set, and test set in a ratio of 7:1.5:1.5;

[0030] According to the task requirements of circuit board defect detection, a deep learning algorithm based on a convolutional neural network model is selected as the basic architecture. A defect recognition model suitable for infrared image defect detection is built. The relationship between the thermal distribution characteristics and the defect pattern is learned to determine whether there are defects. The configured convolutional neural network model and training set are input into the deep learning framework to start model training. During the training process, the parameters are continuously adjusted according to the loss function to gradually learn the relationship between the thermal distribution characteristics and the defect pattern. At the same time, the performance of the model is regularly evaluated using the validation set. After training is completed, the model is finally evaluated using the test set. The model is optimized based on the evaluation results. After multiple optimization and evaluations, the final defect recognition model is obtained until the model performance meets the actual application requirements.

[0031] The trained defect recognition model is deployed in the actual production environment to detect defects in circuit boards in real time. The thermal distribution characteristics are input and the results of whether defects exist are output.

[0032] A further improvement of the technical solution of the present invention is that the process of outputting the result of whether a defect exists is as follows:

[0033] Analyze the thermal distribution features extracted from infrared images, including temperature mean, temperature standard deviation, temperature range, temperature gradient, hot spot area, and hot spot temperature peak value. These features are then organized into the input format required by the defect recognition model. The thermal distribution features of each region are combined into a feature vector. The feature vectors of multiple regions are combined into a feature matrix. The thermal distribution features are then input into the deployed defect recognition model.

[0034] The defect recognition model infers the input thermal distribution features based on the learned relationship between thermal distribution features and defect patterns, and compares the thermal distribution features with known normal feature patterns and defect feature patterns one by one to find out whether there are any abnormalities in the thermal distribution features, and then determine whether there are defects.

[0035] If any thermal distribution feature exceeds the set reference value, the thermal distribution feature is considered to be defective. Based on the result of the defect judgment, a normal or defective classification result is output. If all thermal distribution features are within the set reference value range, a normal classification result is output. If at least one thermal distribution feature exceeds the reference value, a defective classification result is output.

[0036] A further improvement of the technical solution of the present invention is that: S5 specifically includes:

[0037] Input the thermal distribution feature sequence table extracted from the infrared image into the trained defect recognition model, output the result of whether there is a defect, and then determine the presence of abnormal thermal distribution features and preliminarily judge the type of defect;

[0038] Based on the model's inference results, pattern recognition technology is used to locate defects on the circuit board. The spatial information of the thermal distribution characteristics in the infrared image is used to mark the defect area. The specific location of the defect is then further confirmed by comparing abnormal changes in the thermal distribution characteristics.

[0039] Based on the located defect position and the abnormality of the thermal distribution characteristics, the defect is diagnosed, the degree of deviation between the thermal distribution characteristics with abnormal conditions and the reference value is analyzed, the defect severity coefficient is calculated, the severity of the defect is determined, and then the impact of the defect on the performance of the circuit board is evaluated.

[0040] A further improvement of the technical solution of the present invention is that the process of determining the severity of the defect is:

[0041] Based on the located defect location, the thermal distribution characteristic data of the defect location and its surrounding area are extracted from the infrared image. At the same time, the thermal distribution characteristic reference value of the corresponding position when the circuit board is operating normally is obtained. The abnormal data of the defect and the reference value are systematically sorted and a corresponding relationship table is established to ensure that each abnormal characteristic has a clear reference benchmark;

[0042] For each of the collected abnormal thermal distribution features, the degree of deviation from the reference value is calculated one by one. Based on the degree of feature deviation, a weight is assigned to each abnormal thermal distribution feature. The weight value is determined according to the type of circuit board and the working environment. The defect severity coefficient is calculated by combining the deviation and weight of each thermal distribution feature.

[0043] Based on the type of PCB and the working environment, and the impact of defects on PCB performance, different defect levels are pre-set, namely slight defects, moderate defects and severe defects. In addition, the corresponding defect classification threshold is set for each defect level in combination with the defect severity coefficient.

[0044] According to the calculated defect severity coefficient and the divided defect level, the severity of the defect is judged and the level of the defect is determined.

[0045] A further improvement of the technical solution of the present invention is that the calculation process of the defect severity coefficient is:

[0046] Based on the output of the defect recognition model, the specific location of the defect on the circuit board is determined, and the abnormal thermal distribution characteristic data of the defect location and its surrounding area are extracted from the infrared image;

[0047] Obtain reference values ​​of thermal distribution characteristics corresponding to defect locations when the circuit board is operating normally, and assign weights to each abnormal thermal distribution characteristic;

[0048] For each abnormal thermal distribution feature, the absolute difference between it and the reference value is calculated one by one. Then, for each abnormal thermal distribution feature, the ratio of its absolute difference to the reference value is calculated to indicate the degree of deviation of the abnormal value relative to the reference value.

[0049] For each abnormal thermal distribution feature, the logarithmic term of its deviation ratio is calculated, and the linear deviation is converted into a logarithmic form to better reflect the severity of the deviation. Then, the deviation ratio of each abnormal thermal distribution feature and the logarithmic term are multiplied, and the square root is calculated and added to the weight of the abnormal thermal distribution feature to obtain the comprehensive impact of the abnormal thermal distribution feature;

[0050] The combined effects of all abnormal thermal distribution features are added together to obtain the total combined effect, and the total combined effect is divided by the total number of abnormal thermal distribution features to obtain the defect severity coefficient.

[0051] A further improvement of the technical solution of the present invention is that: S6 specifically includes:

[0052] Extract defect diagnosis information from the defect identification model, including defect type, defect location, and defect level, and organize the defect diagnosis information into a structured table to ensure data integrity and accuracy. At the same time, record a detailed description of the defect;

[0053] Generate a detailed defect diagnosis report based on the organized defect diagnosis results, including: basic information of the circuit board (model, test date, etc.), defect list and overall evaluation;

[0054] Use visualization tools to display defect information in an intuitive way, mark the defect location on the infrared image, and use different colors to distinguish the defect levels. Minor defects are marked in green, medium defects are marked in yellow, and severe defects are marked in red.

[0055] The generated defect diagnosis report and visualization images are output to a standard format and archived, and the defect diagnosis report is provided to relevant technical personnel. At the same time, user feedback is collected to optimize the report content and visualization effects to ensure that the report can meet actual application needs and provide a reference for subsequent inspection work.

[0056] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0057] 1. The present invention provides a high-frequency, high-speed circuit board detection method based on infrared image enhancement and pattern recognition. By combining infrared image enhancement and pattern recognition and adopting a non-contact detection method, the physical damage to the circuit board caused by traditional contact probe detection is avoided. This not only protects the integrity of the circuit board, but also extends its service life and reduces the defective rate caused by detection. This is particularly important for high-precision, high-value products such as high-frequency, high-speed circuit boards.

[0058] 2. The present invention provides a high-frequency, high-speed circuit board detection method based on infrared image enhancement and pattern recognition. Through infrared image enhancement technology, it can highlight areas of slight thermal changes, improve image clarity and contrast, and thus achieve high-precision diagnosis of minor defects on circuit boards. Combined with pattern recognition technology, it can accurately identify the defect type and locate the defect position, greatly improving the accuracy and reliability of detection and providing support for the repair and improvement of circuit boards.

[0059] 3. The present invention provides a high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition. The defect recognition model is pre-trained based on the deep learning algorithm, which realizes the automation of the detection process. During the detection process, the collected infrared image only needs to be input into the model to quickly obtain the detection results, which greatly improves the detection efficiency. At the same time, automated detection also reduces the interference of human factors and improves the consistency and stability of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0061] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0062] Figure 2 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0064] Example 1, as Figure 1 、 Figure 2 As shown, the present invention provides a high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition, comprising the following steps:

[0065] S1. Use an infrared thermal imager to perform non-contact infrared image acquisition of the high-frequency and high-speed circuit board that is powered on, and obtain the original infrared image data containing the temperature distribution of each area of ​​the circuit board. Select a high-resolution and high-sensitivity infrared thermal imager, check whether the lens, battery, storage device and other components of the infrared thermal imager are normal, ensure that they can work stably, and build a relatively closed detection environment without strong interference light sources, control the ambient temperature and humidity within an appropriate range, and prevent external factors from affecting the thermal imager imaging. Install temperature and humidity sensors for real-time monitoring. If they do not meet the requirements, use a dehumidifier for adjustment. At the same time, use a standard blackbody source to calibrate the infrared thermal imager to eliminate instrument errors and ensure that the measured temperature data is accurate and reliable. Install the high-frequency and high-speed circuit board to be tested on the test fixture, ensure that the circuit board is firmly installed and will not shake during the test to affect the imaging quality, provide it with power and working signals according to the design requirements of the circuit board, so that the circuit board is in normal working condition, and closely observe the working indicator light, voltage and current parameters of the circuit board during the power-on process to ensure that it is running stably without abnormal heating or faults. For example, after power is turned on, wait for a while for the circuit board to reach a stable working state. This process generally takes a few minutes to more than ten minutes. According to the structure and layout of the circuit board, plan multiple shooting angles to fully cover all areas of the circuit board to avoid missing important parts. Operate the infrared thermal imager to shoot the high-frequency and high-speed circuit board that is powered on from the planned angle. Keep the distance and angle between the thermal imager and the circuit board relatively fixed each time to reduce the image difference caused by changes in shooting conditions. While collecting the image, record the time of image collection, circuit board number and shooting angle information. Store the collected original infrared image data containing the temperature distribution of each area of ​​the circuit board in the data warehouse to facilitate subsequent data processing and analysis. After completing the image collection, check whether the collected original infrared image is complete and whether there is any missing or damaged situation. If the image is found to be incomplete, re-collect the image at that angle to ensure the integrity of the data, and then back up the qualified original infrared image data. At the same time, organize and classify the data according to the circuit board number and shooting time, and establish a data directory.

[0066] S2. Use histogram equalization and filtering denoising algorithms to enhance the infrared image, highlight the area of ​​small thermal changes, and improve the image clarity and contrast. Import the collected original infrared image into the image processing software, view the overall image, understand the characteristics of the temperature distribution of each area of ​​the circuit board, identify whether there is obvious noise interference or insufficient contrast, and preliminarily set the relevant parameters of the histogram equalization and filtering denoising algorithms based on the image characteristics. Use the selected histogram equalization algorithm to process the infrared image. By stretching the grayscale range of the image, the details of the originally darker or brighter areas in the image are made clearer and the area of ​​small thermal changes is highlighted. During the processing, the pixel grayscale value is adjusted according to the grayscale histogram of the image to enhance the contrast of the image. Observe the image effect after histogram equalization processing. If it is found that the regional contrast is too high or too low, adjust the algorithm parameters until the ideal contrast effect is achieved to ensure that the area of ​​small thermal changes can be clearly displayed. According to the preliminarily set filtering algorithm and parameters, the image after histogram equalization processing is processed. Filter denoising: If median filtering is selected, the neighborhood pixel values ​​of each pixel in the image are sorted and the median value is taken as the new value of the pixel, thereby effectively removing salt and pepper noise. If Gaussian filtering is selected, the neighborhood pixels are weighted averaged according to the Gaussian function to eliminate Gaussian noise. The image after filtering and denoising is checked to see if the noise is effectively removed. At the same time, observe whether the edge and detail information of the image are excessively blurred. If the noise removal is not thorough or too much detail is lost, adjust the filter window size or change the filter algorithm and process again until a good denoising effect is achieved. Compare the infrared image after histogram equalization and filtering denoising with the original infrared image and conduct a comprehensive evaluation in terms of clarity, contrast, and noise removal. Check whether the area of ​​small thermal changes is more prominent and whether the temperature distribution characteristics of the circuit board are more obvious. Based on the comprehensive evaluation results, fine-tune the parameters of the histogram equalization and filtering denoising algorithms to further optimize the image processing effect. When a satisfactory enhancement effect is achieved, output and save the processed infrared image.

[0067] S3. Extract thermal distribution features from the enhanced infrared image and obtain a thermal distribution feature sequence list as input data for subsequent pattern recognition, wherein the thermal distribution features include regional temperature mean, regional temperature standard deviation, temperature range, temperature gradient, hot spot area and hot spot temperature peak. Import the enhanced infrared image into the image analysis software, use the clustering algorithm to perform image segmentation, distinguish different areas of the circuit board, further refine the segmentation results based on the circuit board layout and component distribution, divide the key areas of the chip, solder joints and traces, and assign a unique identifier to each area, record the area number and location information, ensure that the subsequent feature extraction can accurately correspond to the specific part, and record the boundary information of each area. For each divided area, extract the basic thermal features including regional temperature mean, regional temperature standard deviation and temperature range, as well as the advanced thermal features including temperature gradient, hot spot area and hot spot temperature peak. For each divided area, count the temperature values ​​of all pixels in the area, calculate the average temperature value of all pixels as the temperature mean of the area, reflecting the overall temperature level of the area, calculate the temperature standard deviation based on the temperature value of the pixels in each area, reflecting the area The degree of dispersion of the temperature values ​​in the area. The larger the standard deviation, the more uneven the temperature distribution in the area, and the presence of local overheating or overcooling. Find the maximum and minimum temperature in each area and calculate the difference, that is, the temperature range. The temperature range can intuitively reflect the temperature variation range in the area and help understand the amplitude of temperature fluctuations. Use the Sobel operator to calculate the temperature gradient of the pixel points in each area. The temperature gradient represents the rate of change of temperature in space and can reflect the trend of temperature distribution. By setting the required temperature threshold, the area in the image with a temperature higher than the temperature threshold is identified as a hot spot, and each is calculated. The area of ​​the hot spot can reflect the size of the abnormal heating area. In each hot spot area, find the pixel with the highest temperature. Its temperature value is the peak temperature of the hot spot. The peak temperature of the hot spot can directly reflect the highest temperature of the abnormal heating point. Summarize the basic thermal characteristics and advanced thermal characteristics data of each area, and integrate the extracted temperature mean, temperature standard deviation, temperature range, temperature gradient, hot spot area and hot spot temperature peak of each area into a thermal distribution feature sequence table. Sort and normalize the thermal distribution feature sequence table according to the order of the circuit board area to ensure that the feature sequence format is consistent and the information is complete;

[0068] S4. Collect a large number of infrared image samples of normal and defective circuit boards, annotate the defect types and locations, and pre-train a defect recognition model based on a deep learning algorithm to enable defect recognition capabilities. Collect a large number of infrared image samples of high-frequency and high-speed circuit boards from actual production environments, simulated test scenarios, and databases. Cover images of circuit boards of different models and working conditions, ensuring that the samples are diverse and fully represent various situations in actual applications, including normal circuit boards and circuit boards with various typical defects. Perform a preliminary screening of the collected image samples, removing samples that are unusable due to poor shooting quality or data corruption. At the same time, check the integrity of the samples to ensure that each image clearly shows the infrared thermal distribution of the circuit board. Classify and store the screened samples into normal and defective categories. Organize technical personnel to develop defect annotation specifications based on the characteristics of circuit board defects and actual needs. Specifically, defects are annotated according to the thermal distribution features in the thermal distribution feature sequence table. Reference values ​​for each thermal distribution feature are set. Use image annotation tools to annotate the infrared image samples of defective circuit boards according to the annotation specifications, identify the defects in the images, and annotate the locations and types of the defects. The annotated sample data sets are sorted according to a 7:1.5:1.5 ratio is divided into training set, validation set and test set. The training set is used for model training, the validation set is used to adjust model parameters and evaluate model performance during training, and the test set is used to finally evaluate the generalization ability of the model. Among them, the reference value can be obtained based on a large amount of normal sample data statistics, or it can be determined by the experience of technicians combined with the accuracy requirements of high-frequency and high-speed circuit board detection. According to the task requirements of circuit board defect detection, a deep learning algorithm based on a convolutional neural network model is selected as the basic architecture to build a defect recognition model suitable for infrared image defect detection. The relationship between the learned thermal distribution characteristics and defect patterns is used to determine whether there are defects. According to the characteristics of the selected model and sample data set, the training parameters including learning rate, batch size and number of training rounds are configured. The learning rate affects the step size of the model parameter update, the batch size determines the number of samples used in each iteration, and the number of training rounds controls the number of times the model is trained on the entire data set. Through multiple experiments and adjustments, the optimal parameter combination is found to improve the training effect and convergence speed of the model. Data enhancement processing is performed on the training set, including random cropping, rotation, and flipping. Adjust brightness and contrast. Through data augmentation, generate more training samples to enable the model to learn more robust features and improve its generalization ability. Input the configured convolutional neural network model and training set into the deep learning framework to begin model training. During training, continuously adjust parameters based on the loss function to gradually learn the relationship between thermal distribution characteristics and defect patterns. Simultaneously, use the validation set to regularly evaluate model performance, observe changes in the loss function and accuracy, and identify problems that arise during model training. After training is complete, use the test set for a final performance evaluation of the model. Evaluation metrics include accuracy, recall, and F1 value. Based on the evaluation results, analyze the model's performance on different defect types and identify its strengths and weaknesses. If the model performance does not meet expectations, optimize the model based on the evaluation results. After multiple optimization and evaluation cycles until the model performance meets actual application requirements, the final defect recognition model is obtained. The trained defect recognition model is deployed in an actual production environment for real-time detection of PCB defects. Input thermal distribution characteristics are used, and output indicates whether a defect exists.

[0069] In addition, the process of outputting the result of whether there is a defect is as follows:

[0070] Analyze the thermal distribution features extracted from the infrared image, including the temperature mean, temperature standard deviation, temperature range, temperature gradient, hot spot area, and hot spot temperature peak value, and organize them into the input format required by the defect recognition model. The thermal distribution features of each area are combined into a feature vector, and the feature vectors of multiple areas are combined into a feature matrix. Then, the thermal distribution features are input into the deployed defect recognition model. The defect recognition model infers the input thermal distribution features based on the relationship between the learned thermal distribution features and the defect patterns, and compares the thermal distribution features with the known normal feature patterns and defect feature patterns one by one to find out whether there are any abnormalities in the thermal distribution features, and then judge whether there are defects. If any thermal distribution feature exceeds the set reference value, it is considered that the thermal distribution feature is defective. Then, based on the result of the defect judgment, a normal or defective classification result is output. If all thermal distribution features are within the set reference value range, a normal classification result is output. If at least one thermal distribution feature exceeds the reference value, a defective classification result is output.

[0071] S5. Input the thermal distribution features in the thermal distribution feature sequence table into the defect recognition model, analyze and classify the extracted features, identify the defect type, and combine the pattern recognition results to accurately locate the position of the defect on the circuit board, diagnose the defect based on the defect type, and evaluate the impact of the defect;

[0072] S6. Output a defect diagnosis report containing defect type and location information, and visualize the identified defect information.

[0073] Example 2, as Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, S5 specifically includes:

[0074] The thermal distribution feature sequence table extracted from the infrared image is input into the trained defect recognition model, and the result of whether there is a defect is output, thereby determining the presence of abnormal thermal distribution features and preliminarily judging the type of defect. Based on the inference results of the model, the position of the defect on the circuit board is located in combination with pattern recognition technology. The spatial information of the thermal distribution features in the infrared image is used to mark the defect area. Then, by comparing the abnormal changes in the thermal distribution features, the specific location of the defect is further confirmed to ensure the positioning accuracy. With the dual verification of the model and image information, the reliability of defect positioning is improved. According to the located defect position and the abnormality of the thermal distribution features, the defect is diagnosed, the degree of deviation between the thermal distribution features with abnormal conditions and the reference value is analyzed, the defect severity coefficient is calculated, the severity of the defect is determined, and the impact of the defect on the performance of the circuit board is evaluated.

[0075] In addition, the process of determining the severity of the defect is:

[0076] Based on the located defect position, the thermal distribution feature data of the defect position and its surrounding area are extracted from the infrared image. At the same time, the thermal distribution feature reference value of the corresponding position when the circuit board is operating normally is obtained. The abnormal data and reference value with defects are systematically sorted out, and a corresponding relationship table is established to ensure that each abnormal feature has a clear reference benchmark. For the collected abnormal thermal distribution features, the degree of deviation from the reference value is calculated one by one. Based on the degree of feature deviation, a weight is assigned to each abnormal thermal distribution feature, wherein the weight value is determined according to the type and working environment of the circuit board. The defect severity coefficient is calculated in combination with the deviation and weight of each thermal distribution feature. According to the type and working environment of the circuit board and the degree of influence of the defect on the performance of the circuit board, different defect levels are pre-set, namely slight defects, medium defects and severe defects, and the corresponding defect classification threshold is divided for each defect level in combination with the defect severity coefficient. According to the calculated defect severity coefficient and the classified defect level, the severity of the defect is judged and the determination level of the defect is determined.

[0077] The calculation process of the defect severity coefficient is:

[0078] According to the output results of the defect recognition model, the specific location of the defect on the circuit board is determined, and the abnormal thermal distribution feature data of the defect location and its surrounding area are extracted from the infrared image. The thermal distribution feature reference value corresponding to the defect location when the circuit board is operating normally is obtained, and a weight is assigned to each abnormal thermal distribution feature. For each abnormal thermal distribution feature, the absolute difference between it and the reference value is calculated one by one. Then, for each abnormal thermal distribution feature, the ratio of its absolute difference to the reference value is calculated to represent the degree of deviation of the abnormal value relative to the reference value. For each abnormal thermal distribution feature, the logarithmic term of its deviation ratio is calculated, and the linear deviation is converted into a logarithmic form to better reflect the severity of the deviation. Then, the deviation ratio of each abnormal thermal distribution feature and the logarithmic term are multiplied, and the square root is calculated. The square root is multiplied by the weight of the abnormal thermal distribution feature to obtain the comprehensive impact of the abnormal thermal distribution feature. The comprehensive impact of all abnormal thermal distribution features is added to obtain the total comprehensive impact, and the total comprehensive impact is divided by the total number of abnormal thermal distribution features to obtain the defect severity coefficient.

[0079] The calculation expression of defect severity coefficient is:

[0080]

[0081] Where S is the defect severity coefficient, which reflects the comprehensive severity of the defect, N is the total number of abnormal thermal distribution features, and w i is the weight of the ith abnormal thermal distribution feature, which is set according to the circuit board type and working environment, and reflects the impact of the abnormal thermal distribution feature on the severity of the defect. iis the actual value of the ith abnormal thermal distribution feature, that is, the actual measurement value extracted from the defect position, R i is the reference value of the ith abnormal thermal distribution feature, that is, the standard value when the circuit board is operating normally, |T i -R i | is the deviation degree of the i-th abnormal thermal distribution feature, which represents the absolute difference between the actual value of the abnormal thermal distribution feature and the reference value. When the deviation of all abnormal thermal distribution features is small, S approaches 0, indicating that the defect is not serious;

[0082] Multiple defect levels correspond to multiple defect classification thresholds one by one, and the corresponding relationship is as follows:

[0083] The defect classification threshold for minor defects is: 0 <S≤S L ;

[0084] The defect classification threshold for medium defects is: S L <S≤S H ;

[0085] The threshold for severe defects is: S>S H ;

[0086] Among them, S is the defect severity coefficient, S L is the upper threshold of slight defects and the lower threshold of medium defects, S H is the upper threshold of medium defects and the lower threshold of severe defects, S L =0.3, S H =0.5;

[0087] S6 specifically includes:

[0088] Extract defect diagnosis information from the defect recognition model, including defect type, defect location and defect level, organize the defect diagnosis information into a structured table to ensure the integrity and accuracy of the data, and record a detailed description of the defect. Based on the organized defect diagnosis results, generate a detailed defect diagnosis report, including: basic information of the circuit board (model, test date, etc.), defect list and overall evaluation. The defect list lists the type, location, defect level and description of each defect in detail. The overall evaluation is used to summarize the overall health status of the circuit board and the recommended maintenance measures. Use visualization tools (Matplotlib, Plotly or circuit board inspection software) to visualize the defects. The information is displayed in an intuitive manner, and the defect location is marked on the infrared image. Different colors are used to distinguish the defect levels. Minor defects are marked in green, medium defects are marked in yellow, and severe defects are marked in red. At the same time, a thermal map of temperature distribution is added to highlight thermal anomaly areas, allowing technicians to quickly locate and evaluate defects. The generated defect diagnosis report and visualization image are output in a standard format and archived. The defect diagnosis report is provided to relevant technicians so that corresponding maintenance or improvement measures can be taken according to the report content. At the same time, user feedback is collected to optimize the report content and visualization effects to ensure that the report can meet actual application needs and provide a reference for subsequent detection work.

[0089] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition, characterized in that: The following steps are involved: S1. Use an infrared thermal imager to perform non-contact infrared image acquisition on a high-frequency, high-speed circuit board that is powered on, and obtain raw infrared image data containing the temperature distribution of each area of ​​the circuit board; S2. Use histogram equalization and filtering denoising algorithms to enhance the infrared image and highlight the area with small thermal changes; S3, extracting thermal distribution features from the enhanced infrared image to obtain a thermal distribution feature sequence list; S4. Collect a large number of infrared image samples of normal and defective circuit boards, annotate the defect types and locations, and pre-train a defect recognition model based on a deep learning algorithm; S5. Input the thermal distribution features in the thermal distribution feature sequence table into the defect recognition model, analyze and classify the extracted features, identify the defect type, and locate the position of the defect on the circuit board based on the pattern recognition results. Perform a diagnosis based on the defect type and evaluate the impact of the defect. S6. Output a defect diagnosis report containing defect type and location information, and visualize the identified defect information.

2. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 1 is characterized in that: Said S1 specifically includes: Select an infrared thermal imager and build a relatively closed detection environment without strong interference light sources. At the same time, use a standard blackbody source to calibrate the infrared thermal imager to eliminate instrument errors. Install the high-frequency and high-speed circuit board to be tested on the test fixture. Provide it with power and working signals according to the design requirements of the circuit board to put it in normal working state. After powering on, wait for a period of time for the circuit board to reach a stable working state. Based on the structure and layout of the circuit board, multiple shooting angles are planned to cover various areas of the circuit board. The infrared thermal imager is used to shoot the high-frequency and high-speed circuit board while it is powered on from the planned angles. While capturing the image, the image capture time, circuit board number, and shooting angle are recorded. The collected raw infrared image data containing the temperature distribution of each area of ​​the circuit board is stored in the data warehouse. After completing the image acquisition, check whether the original infrared image is complete. If the image is found to be incomplete, re-acquire the image at that angle and then back up the original infrared image data that has passed the inspection. At the same time, organize and classify the data according to the circuit board number and shooting time to establish a data directory.

3. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 1 is characterized in that: The S2 specifically includes: Import the collected original infrared image into the image processing software to check the overall image situation and preliminarily set the relevant parameters of the histogram equalization and filtering denoising algorithm according to the image characteristics; The infrared image is processed using the selected histogram equalization algorithm to highlight the area with slight thermal changes by stretching the grayscale range of the image, and the pixel grayscale value is adjusted according to the grayscale histogram of the image to enhance the contrast of the image; Perform filtering and denoising on the image after histogram equalization according to the initially set filtering algorithm and parameters. Check the filtered and denoised image to see if the noise is effectively removed and whether the edges and details of the image are overly blurred. The infrared image after histogram equalization and filtering denoising is compared with the original infrared image, and a comprehensive evaluation is performed from multiple aspects such as clarity, contrast and noise removal. Based on the comprehensive evaluation results, the parameters of the histogram equalization and filtering denoising algorithms are fine-tuned to further optimize the image processing effect. When a satisfactory enhancement effect is achieved, the processed infrared image is output and saved.

4. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 1 is characterized in that: The S3 specifically includes: Import the enhanced infrared image into the image analysis software and use the clustering algorithm to segment the image and distinguish different areas of the circuit board. Based on the circuit board layout and component distribution, further refine the segmentation results to divide the key areas of chips, solder joints and traces. Each area is given a unique identifier, and the area number and location information are recorded. At the same time, the boundary information of each area is recorded. For each divided area, basic thermal features including regional temperature mean, regional temperature standard deviation and temperature range are extracted, as well as advanced thermal features including temperature gradient, hot spot area and hot spot temperature peak; The basic thermal characteristics and advanced thermal characteristics data of each area are summarized, and the extracted temperature mean, temperature standard deviation, temperature range, temperature gradient, hot spot area and hot spot temperature peak characteristics of each area are integrated into a thermal distribution feature sequence table, and the thermal distribution feature sequence table is sorted and normalized according to the order of the circuit board areas.

5. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 4 is characterized in that: The S4 specifically includes: Collect a large number of infrared image samples of high-frequency and high-speed circuit boards from actual production environments, simulated test scenarios, and databases, including normal circuit boards and circuit boards with various typical defects. Perform preliminary screening on the collected image samples and classify and store them according to normal and defective samples. Based on the characteristics of circuit board defects and actual needs, a defect annotation specification is developed. That is, defects are annotated according to the thermal distribution characteristics of the thermal distribution feature sequence table, reference values ​​for each thermal distribution feature are set, and image annotation tools are used to annotate infrared image samples of defective circuit boards according to the annotation specifications. Defects in the image are identified, and the location and type of the defects are marked. The annotated sample data set is divided into a training set, a validation set, and a test set. According to the task requirements of circuit board defect detection, a deep learning algorithm based on a convolutional neural network model is selected as the basic architecture. A defect recognition model suitable for infrared image defect detection is built. The relationship between the thermal distribution characteristics and the defect pattern is learned to determine whether there are defects. The configured convolutional neural network model and training set are input into the deep learning framework to start model training. During the training process, the parameters are continuously adjusted according to the loss function to gradually learn the relationship between the thermal distribution characteristics and the defect pattern. At the same time, the performance of the model is regularly evaluated using the validation set. After training is completed, the model is finally evaluated using the test set. The model is optimized based on the evaluation results. After multiple optimization and evaluations, the final defect recognition model is obtained until the model performance meets the actual application requirements. The trained defect recognition model is deployed in the actual production environment to detect defects in circuit boards in real time. The thermal distribution characteristics are input and the results of whether defects exist are output.

6. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 5 is characterized in that: The process of outputting the result of whether there is a defect is as follows: Analyze the thermal distribution features extracted from infrared images, including temperature mean, temperature standard deviation, temperature range, temperature gradient, hot spot area, and hot spot temperature peak value. These features are then organized into the input format required by the defect recognition model. The thermal distribution features of each region are combined into a feature vector. The feature vectors of multiple regions are combined into a feature matrix. The thermal distribution features are then input into the deployed defect recognition model. The defect recognition model infers the input thermal distribution features based on the learned relationship between thermal distribution features and defect patterns, and compares the thermal distribution features with known normal feature patterns and defect feature patterns one by one to find out whether there are any abnormalities in the thermal distribution features, and then determine whether there are defects. If any thermal distribution feature exceeds the set reference value, the thermal distribution feature is considered to be defective. Based on the result of the defect judgment, a normal or defective classification result is output. If all thermal distribution features are within the set reference value range, a normal classification result is output. If at least one thermal distribution feature exceeds the reference value, a defective classification result is output.

7. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 6, characterized in that: The S5 specifically includes: Input the thermal distribution feature sequence table extracted from the infrared image into the trained defect recognition model, output the result of whether there is a defect, and then determine the presence of abnormal thermal distribution features and preliminarily judge the type of defect; Based on the model's inference results, pattern recognition technology is used to locate defects on the circuit board. The spatial information of the thermal distribution characteristics in the infrared image is used to mark the defect area. The specific location of the defect is then further confirmed by comparing abnormal changes in the thermal distribution characteristics. Based on the located defect position and the abnormality of the thermal distribution characteristics, the defect is diagnosed, the degree of deviation between the thermal distribution characteristics with abnormal conditions and the reference value is analyzed, the defect severity coefficient is calculated, the severity of the defect is determined, and then the impact of the defect on the performance of the circuit board is evaluated.

8. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 7 is characterized in that: The process of determining the severity of the defect is as follows: Based on the located defect location, the thermal distribution characteristic data of the defect location and its surrounding area are extracted from the infrared image. At the same time, the thermal distribution characteristic reference value of the corresponding position when the circuit board is operating normally is obtained. The abnormal data of the defect and the reference value are systematically sorted and a corresponding relationship table is established; For the collected abnormal thermal distribution features, the degree of deviation from the reference value is calculated one by one. Based on the degree of feature deviation, a weight is assigned to each abnormal thermal distribution feature. The defect severity coefficient is calculated by combining the deviation and weight of each thermal distribution feature. Based on the type of PCB and the working environment, and the impact of defects on PCB performance, different defect levels are pre-set, namely slight defects, moderate defects and severe defects. In addition, the corresponding defect classification threshold is set for each defect level in combination with the defect severity coefficient. According to the calculated defect severity coefficient and the divided defect level, the severity of the defect is judged and the level of the defect is determined.

9. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 8, characterized in that: The calculation process of the defect severity coefficient is: Based on the output of the defect recognition model, the specific location of the defect on the circuit board is determined, and the abnormal thermal distribution characteristic data of the defect location and its surrounding area are extracted from the infrared image; Obtain reference values ​​of thermal distribution characteristics corresponding to defect locations when the circuit board is operating normally, and assign weights to each abnormal thermal distribution characteristic; For each abnormal thermal distribution feature, calculate the absolute difference between it and the reference value one by one, and then calculate the ratio of its absolute difference to the reference value for each abnormal thermal distribution feature; For each abnormal thermal distribution feature, calculate the logarithmic term of its deviation ratio, convert the linear deviation into logarithmic form, and then multiply the deviation ratio of each abnormal thermal distribution feature by the logarithmic term, calculate the square root, and add it to the weight of the abnormal thermal distribution feature to obtain the comprehensive impact of the abnormal thermal distribution feature; The combined effects of all abnormal thermal distribution features are added together to obtain the total combined effect, and the total combined effect is divided by the total number of abnormal thermal distribution features to obtain the defect severity coefficient.

10. The high-frequency and high-speed circuit board detection method based on infrared image enhancement and pattern recognition according to claim 8, characterized in that: The S6 specifically includes: Extract defect diagnosis information from the defect recognition model, including defect type, defect location, and defect level, organize the defect diagnosis information into a structured table, and record a detailed description of the defect; Generate a detailed defect diagnosis report based on the organized defect diagnosis results, including: basic information of the circuit board, defect list and overall evaluation; Use visualization tools to display defect information in an intuitive way, mark the defect location on the infrared image, and use different colors to distinguish the defect levels. Minor defects are marked in green, medium defects are marked in yellow, and severe defects are marked in red. The generated defect diagnosis report and visualization images are output to a standard format and archived, and the defect diagnosis report is provided to relevant technical personnel. At the same time, user feedback is collected to optimize the report content and visualization effects.

Images