A test analysis method, system and storage medium based on deep learning

Abstract

The present invention discloses a test analysis method, system and storage medium based on deep learning, which obtains historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data; the historical test image data is subjected to image preprocessing and imported into an adversarial neural network model based on deep learning for generating and analyzing training data, and training data is generated in combination with a preset circuit test plan; the training data is imported into a CNN-based image recognition model for training, and real-time test image data of a target circuit board is obtained in real time; the real-time test image data is imported into an image recognition model for abnormal area recognition, and a test report is generated, and the target circuit board is evaluated and analyzed based on the test report to obtain processing evaluation data and processing scheme correction information. The present invention can realize real-time, rapid and accurate detection of circuit board defects, thereby achieving the purpose of rapidly improving the quality of finished products of circuit boards.

Description

Technical Field

[0001] The present invention relates to the field of deep learning and image analysis, and more specifically, to a test analysis method, system and storage medium based on deep learning. Background Art

[0002] In the process of detecting defects in printed circuit boards, it is difficult to achieve intelligent automatic testing and efficient testing due to the limitations of existing technologies. Most tests rely on manpower, making it difficult to quickly improve the quality of finished circuit boards, and the testing process is time-consuming and labor-intensive. Therefore, an efficient testing and analysis method is urgently needed. Summary of the invention

[0003] The present invention overcomes the defects of the prior art and proposes a test analysis method, system and storage medium based on deep learning.

[0004] A first aspect of the present invention provides a test analysis method based on deep learning, comprising:

[0005] Acquire historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data;

[0006] The historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan;

[0007] Importing the training data into a CNN-based image recognition model for training, and acquiring real-time test image data of the target circuit board in real time;

[0008] The real-time test image data is imported into an image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing plan correction information.

[0009] In this solution, the historical test image data of the preset circuit board is obtained, and the test image data includes high-definition image data and thermal imaging image data, specifically:

[0010] Acquire historical test image data of a preset circuit board, and divide the historical test image data into high-definition image data and thermal imaging image data;

[0011] Based on the design information of the preset circuit board, obtain the circuit board component area information;

[0012] Based on the circuit board component area information, the high-definition image data and the thermal imaging image data are associated between the image areas.

[0013] In this solution, the historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, which includes:

[0014] Perform image denoising, smoothing and standardization preprocessing on historical test image data;

[0015] Based on the circuit board component area information, the thermal imaging image data is divided into regions to obtain the circuit area and the component area;

[0016] Based on the circuit area and the component area, color features are extracted from the thermal imaging image data to obtain circuit color features and component color features;

[0017] Perform temperature range analysis on the component area based on the thermal imaging image data to obtain the temperature variation range, and divide N temperature span ranges based on the temperature variation range;

[0018] Based on the N temperature span ranges, contour feature extraction is performed on the component region of the thermal imaging image data to obtain N contour feature data, each contour feature data corresponding to a temperature span range.

[0019] In this solution, the historical test image data is subjected to image preprocessing and imported into the adversarial neural network model based on deep learning to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, specifically:

[0020] Build an adversarial neural network model based on deep learning and mark it as an adversarial generative model; build a generator and a discriminator in the adversarial generative model;

[0021] Import the thermal imaging image data in the historical test image data into the adversarial generation model and use it as real data for adversarial training;

[0022] Vectorize the line color features, component color features, and N contour feature data to generate feature vector data, and import the feature vector data into the generator as a generator model parameter;

[0023] Perform cyclic adversarial training based on the generator and the discriminator until the generator and the discriminator reach Nash equilibrium;

[0024] A preset amount of thermal imaging simulation feature data is generated based on the trained adversarial generative model, and the thermal imaging simulation feature data is used as training data.

[0025] In this solution, the training data is imported into the CNN image recognition model for training, specifically:

[0026] Build a CNN-based image recognition model;

[0027] Importing thermal imaging image data in the historical test image data into the image recognition model to perform a model training and achieve a first preset recognition rate;

[0028] The training data is imported into the image recognition model for secondary model training to achieve a second preset recognition rate.

[0029] In this solution, the real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing solution correction information, specifically:

[0030] Acquire real-time test image data of the target circuit board in real time during N consecutive test time periods, wherein the real-time test image data includes real-time high-definition image data and real-time thermal imaging image data;

[0031] The real-time thermal imaging image data is subjected to image noise reduction, smoothing, and standardization preprocessing and imported into an image recognition model for abnormality recognition based on thermal imaging;

[0032] According to the circuit area and the component area, the circuit abnormal area and the component abnormal area are respectively identified;

[0033] According to the real-time analysis of line abnormal areas, component abnormal areas and N continuous test time periods, the number of abnormal areas and abnormal duration in each continuous test time period are analyzed;

[0034] Generate test reports based on the number of abnormal areas, abnormal duration, line abnormal areas and component abnormal areas.

[0035] In this solution, the real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing solution correction information, specifically:

[0036] Based on the test report, the circuit area and component area of ​​the target circuit board are processed and evaluated to generate processing evaluation data;

[0037] Based on the processing evaluation data, the processing plan adjustment analysis and circuit board design correction analysis are performed to obtain processing plan correction information.

[0038] The second aspect of the present invention further provides a test analysis system based on deep learning, the system comprising: a memory, a processor, the memory comprising a test analysis program based on deep learning, the test analysis program based on deep learning being executed by the processor to implement the following steps:

[0039] Acquire historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data;

[0040] The historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan;

[0041] Importing the training data into a CNN-based image recognition model for training, and acquiring real-time test image data of the target circuit board in real time;

[0042] The real-time test image data is imported into an image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing plan correction information.

[0043] In this solution, the historical test image data of the preset circuit board is obtained, and the test image data includes high-definition image data and thermal imaging image data, specifically:

[0044] Acquire historical test image data of a preset circuit board, and divide the historical test image data into high-definition image data and thermal imaging image data;

[0045] Based on the design information of the preset circuit board, obtain the circuit board component area information;

[0046] Based on the circuit board component area information, the high-definition image data and the thermal imaging image data are associated between the image areas.

[0047] The third aspect of the present invention also provides a computer-readable storage medium, which includes a deep learning-based test analysis program. When the deep learning-based test analysis program is executed by a processor, it implements the steps of the deep learning-based test analysis method as described in any one of the above items.

[0048] The present invention discloses a test analysis method, system and storage medium based on deep learning, which obtains historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data; the historical test image data is subjected to image preprocessing and imported into an adversarial neural network model based on deep learning for generating and analyzing training data, and training data is generated in combination with a preset circuit test plan; the training data is imported into a CNN-based image recognition model for training, and real-time test image data of a target circuit board is obtained in real time; the real-time test image data is imported into an image recognition model for abnormal area recognition, and a test report is generated, and the target circuit board is evaluated and analyzed based on the test report to obtain processing evaluation data and processing scheme correction information. The present invention can realize real-time, rapid and accurate detection of circuit board defects, thereby achieving the purpose of rapidly improving the quality of finished products of circuit boards. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a test analysis method based on deep learning of the present invention is shown;

[0050] Figure 2 A flowchart of the association of high-definition and thermal imaging image data of the present invention is shown;

[0051] Figure 3 The CNN image recognition model training flow chart of the present invention is shown;

[0052] Figure 4 A block diagram of a deep learning-based test analysis system of the present invention is shown. DETAILED DESCRIPTION

[0053] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0055] Figure 1 A flow chart of a test analysis method based on deep learning of the present invention is shown.

[0056] like Figure 1 As shown, the first aspect of the present invention provides a test analysis method based on deep learning, comprising:

[0057] S102, obtaining historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data;

[0058] S104, performing image preprocessing on the historical test image data and importing the image data into a deep learning-based adversarial neural network model to generate and analyze training data, and generating training data in combination with a preset circuit test plan;

[0059] S106, importing the training data into a CNN-based image recognition model for training, and acquiring real-time test image data of the target circuit board in real time;

[0060] S108, importing the real-time test image data into an image recognition model to identify abnormal areas, and generating a test report, evaluating and analyzing the target circuit board based on the test report, and obtaining processing evaluation data and processing solution correction information.

[0061] It should be noted that the test report is a report generated by performing a real-time power-on test. The test specifically involves power-on and functional testing of the circuit board. The specific test content is set by the designer.

[0062] Figure 2 A flow chart of associating high-definition and thermal imaging image data according to the present invention is shown.

[0063] According to an embodiment of the present invention, the acquisition of historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data, is specifically:

[0064] S202, obtaining historical test image data of a preset circuit board, and dividing the historical test image data into high-definition image data and thermal imaging image data;

[0065] S204, obtaining circuit board component area information based on preset circuit board design information;

[0066] S206, based on the circuit board component area information, the high-definition image data and the thermal imaging image data are associated with each other in terms of image areas.

[0067] It should be noted that, through the association between image areas, the location area corresponding to another image can be quickly obtained based on a certain position of a certain image. For example, if a temperature anomaly appears in a certain area of ​​the thermal image, the corresponding area of ​​the high-definition image can be retrieved based on the association and the corresponding component or circuit can be identified. The circuit board component area information includes each component area and circuit area information.

[0068] According to an embodiment of the present invention, the historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, which includes:

[0069] Perform image denoising, smoothing and standardization preprocessing on historical test image data;

[0070] Based on the circuit board component area information, the thermal imaging image data is divided into regions to obtain the circuit area and the component area;

[0071] Based on the circuit area and the component area, color features are extracted from the thermal imaging image data to obtain circuit color features and component color features;

[0072] Perform temperature range analysis on the component area based on the thermal imaging image data to obtain the temperature variation range, and divide N temperature span ranges based on the temperature variation range;

[0073] Based on the N temperature span ranges, contour feature extraction is performed on the component region of the thermal imaging image data to obtain N contour feature data, each contour feature data corresponding to a temperature span range.

[0074] It should be noted that the temperature range analysis of the component area based on the thermal imaging image data is performed to obtain the temperature change range. In the N temperature span ranges divided based on the temperature change range, the temperature change range is specifically a larger temperature span range, there is a minimum value and a maximum value, and the N temperature span ranges are specifically divided into intervals between the minimum value and the maximum value, and the interval span can be set by the user, and the N is generally 2 to 3, that is, generally 3 span ranges, and 3 contour feature data are obtained, corresponding to the core temperature area, the middle area, and the edge area of ​​the component, respectively. The closer to the core temperature area, the higher the temperature. The present invention can extract the contours of different areas by dividing different temperature span areas, and further analyze its contour features and import the GAN (adversarial neural network model) model to generate relevant features and obtain training data, thereby improving the CNN model's training and recognition effect on abnormal areas in the image, and further realizing accurate circuit board automatic detection. The color feature can effectively reflect the temperature change characteristics corresponding to the thermal imaging and the corresponding temperature value. The thermal imaging image data includes data obtained from multiple test cycles. The N temperature span ranges, circuit color characteristics, component color characteristics, and contour feature data are obtained by analysis within a certain cycle. Different circuit color characteristics, component color characteristics, and contour feature data can be obtained in different cycles.

[0075] According to an embodiment of the present invention, the historical test image data is subjected to image preprocessing and imported into an adversarial neural network model based on deep learning to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, specifically:

[0076] Build an adversarial neural network model based on deep learning and label it as an adversarial generative model;

[0077] Building generators and discriminators in adversarial generative models;

[0078] Import the thermal imaging image data in the historical test image data into the adversarial generation model and use it as real data for adversarial training;

[0079] Vectorize the line color features, component color features, and N contour feature data to generate feature vector data, and import the feature vector data into the generator as a generator model parameter;

[0080] Perform cyclic adversarial training based on the generator and the discriminator until the generator and the discriminator reach Nash equilibrium;

[0081] A preset amount of thermal imaging simulation feature data is generated based on the trained adversarial generative model, and the thermal imaging simulation feature data is used as training data.

[0082] It should be noted that in the adversarial generation model, the present invention mainly generates feature data based on thermal imaging. This embodiment analyzes the color features and contour features of the circuit board during testing in the historical image data, and further combines the GAN model to simulate the generation of feature data, thereby generating more realistic training data, improving the subsequent image recognition model's recognition of normal and abnormal areas of thermal imaging, and improving test efficiency.

[0083] In addition, the adversarial generative model is based on N contour feature data and related color features extracted from historical thermal imaging data, and generates feature data through a generator. It can simulate the generation of thermal imaging feature data within a certain temperature span and obtain simulated data. By importing the simulated data into the training of the image recognition model, it can more accurately identify the subsequent real-time test image data and identify abnormal areas.

[0084] It is worth mentioning that the present invention performs thermal imaging anomaly analysis and identification in the form of regions, instead of using a single point as an analysis method. This enables anomaly identification to be performed based on regionality, improves the accuracy of anomaly identification, and realizes rapid anomaly monitoring and abnormal area extraction of the global image of the circuit board. In the subsequent process, circuit board defect correction or processing plan correction can also be performed in the corresponding abnormal area.

[0085] Figure 3The flowchart of CNN image recognition model training of the present invention is shown.

[0086] According to an embodiment of the present invention, the importing of the training data into a CNN-based image recognition model for training is specifically as follows:

[0087] S302, constructing an image recognition model based on CNN;

[0088] S304, importing the thermal imaging image data in the historical test image data into the image recognition model to perform a model training, and achieving a first preset recognition rate;

[0089] S306, importing the training data into the image recognition model for secondary model training, and achieving a second preset recognition rate.

[0090] It should be noted that the second preset recognition rate is greater than the first preset recognition rate.

[0091] According to an embodiment of the present invention, the real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing solution correction information, specifically:

[0092] Acquire real-time test image data of the target circuit board in real time during N consecutive test time periods, wherein the real-time test image data includes real-time high-definition image data and real-time thermal imaging image data;

[0093] The real-time thermal imaging image data is subjected to image noise reduction, smoothing, and standardization preprocessing and imported into an image recognition model for abnormality recognition based on thermal imaging;

[0094] According to the circuit area and the component area, the circuit abnormal area and the component abnormal area are respectively identified;

[0095] According to the real-time analysis of line abnormal areas, component abnormal areas and N continuous test time periods, the number of abnormal areas and abnormal duration in each continuous test time period are analyzed;

[0096] Generate test reports based on the number of abnormal areas, abnormal duration, line abnormal areas and component abnormal areas.

[0097] It should be noted that the test report includes abnormal reports of various areas of the target circuit board in multiple continuous test time periods, which can effectively reflect the temperature abnormality areas after a test and adjust the plan based on the test report. The test time period is generally a functional test with power on.

[0098] In addition, through the regional image model analysis of the present invention, global temperature thermal imaging analysis can be performed on the entire circuit board, and it is not limited to the temperature abnormality monitoring and detection of a certain point. For example, the temperature abnormality of a component affects the temperature of the remaining adjacent areas, causing the temperature of the remaining adjacent areas to fluctuate to a certain extent. Through the present invention, the relevant abnormal temperature control areas can be identified and marked in real time, further realizing the intelligent and automated testing process of the circuit board.

[0099] According to an embodiment of the present invention, the real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing solution correction information, specifically:

[0100] Based on the test report, the circuit area and component area of ​​the target circuit board are processed and evaluated to generate processing evaluation data;

[0101] Based on the processing evaluation data, the processing plan adjustment analysis and circuit board design correction analysis are performed to obtain processing plan correction information.

[0102] It should be noted that the processing evaluation data includes processing evaluation and design rationality evaluation data for circuits and components.

[0103] According to an embodiment of the present invention, it also includes:

[0104] Based on a preset test cycle and in combination with a test plan, M subdivision cycles are divided, and the division is a continuous division based on time;

[0105] The target circuit board is tested based on a preset test cycle, and M pieces of thermal imaging real-time data corresponding to M subdivided cycles are obtained in real time;

[0106] Based on the system database, retrieve and obtain the standard test thermal imaging data within M subdivision periods;

[0107] Importing the standard test thermal imaging data within the M subdivided periods into an image recognition model for feature recognition and extraction to obtain M standard feature data;

[0108] The standard feature data includes color features and thermal imaging contour features of the circuit area and component area;

[0109] Importing the M pieces of thermal imaging real-time data into an image recognition model for feature recognition and extraction to obtain M pieces of real-time feature data;

[0110] The real-time feature data includes color features and thermal imaging contour features of the circuit area and component area;

[0111] Taking each segmentation period as the analysis unit, M real-time feature data and M standard feature data are compared and analyzed, and the similarities of M features are calculated based on the Mahalanobis distance;

[0112] Based on a preset test cycle and M sets of real-time thermal imaging data, the temperature fluctuation data of the target circuit board test process is obtained;

[0113] The temperature fluctuation data includes temperature fluctuation curves of the circuit area and the component area;

[0114] Acquire standard temperature fluctuation data based on standard test thermal imaging data;

[0115] Based on the temperature fluctuation data and the standard temperature fluctuation data, a periodic temperature difference analysis based on M periods is performed to obtain M temperature control difference degrees;

[0116] Based on M feature similarities and M temperature control differences, life and quality assessment of the target circuit board are performed to obtain life assessment information and quality assessment information.

[0117] It should be noted that the preset test cycle is a longer cycle, and refined analysis is performed by dividing each time period. The test plan includes the test duration and test intensity of different components in different time periods within a preset test cycle. The M feature similarities and M temperature control differences can reflect the test differences between the target circuit board and the standard production standard finished product. The higher the feature similarity or the lower the temperature control difference, the better the temperature control of the target circuit board, the higher the quality of the finished product, and the higher the corresponding finished product life. The feature similarity and temperature control difference are respectively used to evaluate the quality of the target circuit board in the regional feature dimension and the temperature curve change dimension, realizing a dual-dimensional circuit board abnormality analysis. It is worth mentioning that through the dual-dimensional analysis, the small temperature control differences in the circuit board can be amplified. The difference analysis can be realized, and the precise and accurate testing process can be further realized.

[0118] Figure 4 A block diagram of a deep learning-based test analysis system of the present invention is shown.

[0119] The second aspect of the present invention further provides a deep learning-based test analysis system 4, which includes: a memory 41 and a processor 42, wherein the memory includes a deep learning-based test analysis program, and when the deep learning-based test analysis program is executed by the processor, the following steps are implemented:

[0120] Acquire historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data;

[0121] The historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan;

[0122] Importing the training data into a CNN-based image recognition model for training, and acquiring real-time test image data of the target circuit board in real time;

[0123] The real-time test image data is imported into an image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing plan correction information.

[0124] It should be noted that the test report is a report generated by performing a real-time power-on test. The test specifically involves power-on and functional testing of the circuit board. The specific test content is set by the designer.

[0125] According to an embodiment of the present invention, the acquisition of historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data, is specifically:

[0126] Acquire historical test image data of a preset circuit board, and divide the historical test image data into high-definition image data and thermal imaging image data;

[0127] Based on the design information of the preset circuit board, obtain the circuit board component area information;

[0128] Based on the circuit board component area information, the high-definition image data and the thermal imaging image data are associated between the image areas.

[0129] It should be noted that, through the association between image areas, the location area corresponding to another image can be quickly obtained based on a certain position of a certain image. For example, if a temperature anomaly appears in a certain area of ​​the thermal image, the corresponding area of ​​the high-definition image can be retrieved based on the association and the corresponding component or circuit can be identified. The circuit board component area information includes each component area and circuit area information.

[0130] According to an embodiment of the present invention, the historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, which includes:

[0131] Perform image denoising, smoothing and standardization preprocessing on historical test image data;

[0132] Based on the circuit board component area information, the thermal imaging image data is divided into regions to obtain the circuit area and the component area;

[0133] Based on the circuit area and the component area, color features are extracted from the thermal imaging image data to obtain circuit color features and component color features;

[0134] Perform temperature range analysis on the component area based on the thermal imaging image data to obtain the temperature variation range, and divide N temperature span ranges based on the temperature variation range;

[0135] Based on the N temperature span ranges, contour feature extraction is performed on the component region of the thermal imaging image data to obtain N contour feature data, each contour feature data corresponding to a temperature span range.

[0136] It should be noted that the temperature range analysis of the component area based on the thermal imaging image data is performed to obtain the temperature change range. In the N temperature span ranges divided based on the temperature change range, the temperature change range is specifically a larger temperature span range, there is a minimum value and a maximum value, and the N temperature span ranges are specifically divided into intervals between the minimum value and the maximum value, and the interval span can be set by the user, and the N is generally 2 to 3, that is, generally 3 span ranges, and 3 contour feature data are obtained, corresponding to the core temperature area, the middle area, and the edge area of ​​the component, respectively. The closer to the core temperature area, the higher the temperature. The present invention can extract the contours of different areas by dividing different temperature span areas, and further analyze its contour features and import the GAN (adversarial neural network model) model to generate relevant features and obtain training data, thereby improving the CNN model's training and recognition effect on abnormal areas in the image, and further realizing accurate circuit board automatic detection. The color feature can effectively reflect the temperature change characteristics corresponding to the thermal imaging and the corresponding temperature value. The thermal imaging image data includes data obtained from multiple test cycles. The N temperature span ranges, circuit color characteristics, component color characteristics, and contour feature data are obtained by analysis within a certain cycle. Different circuit color characteristics, component color characteristics, and contour feature data can be obtained in different cycles.

[0137] According to an embodiment of the present invention, the historical test image data is subjected to image preprocessing and imported into an adversarial neural network model based on deep learning to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, specifically:

[0138] Build an adversarial neural network model based on deep learning and mark it as an adversarial generative model; build a generator and a discriminator in the adversarial generative model;

[0139] Import the thermal imaging image data in the historical test image data into the adversarial generation model and use it as real data for adversarial training;

[0140] Vectorize the line color features, component color features, and N contour feature data to generate feature vector data, and import the feature vector data into the generator as a generator model parameter;

[0141] Perform cyclic adversarial training based on the generator and the discriminator until the generator and the discriminator reach Nash equilibrium;

[0142] A preset amount of thermal imaging simulation feature data is generated based on the trained adversarial generative model, and the thermal imaging simulation feature data is used as training data.

[0143] It should be noted that in the adversarial generation model, the present invention mainly generates feature data based on thermal imaging. This embodiment analyzes the color features and contour features of the circuit board during testing in the historical image data, and further combines the GAN model to simulate the generation of feature data, thereby generating more realistic training data, improving the subsequent image recognition model's recognition of normal and abnormal areas of thermal imaging, and improving test efficiency.

[0144] In addition, the adversarial generative model is based on N contour feature data and related color features extracted from historical thermal imaging data, and generates feature data through a generator. It can simulate the generation of thermal imaging feature data within a certain temperature span and obtain simulated data. By importing the simulated data into the training of the image recognition model, it can more accurately identify the subsequent real-time test image data and identify abnormal areas.

[0145] It is worth mentioning that the present invention performs thermal imaging anomaly analysis and identification in the form of regions, instead of using a single point as an analysis method. This enables anomaly identification to be performed based on regionality, improves the accuracy of anomaly identification, and realizes rapid anomaly monitoring and abnormal area extraction of the global image of the circuit board. In the subsequent process, circuit board defect correction or processing plan correction can also be performed in the corresponding abnormal area.

[0146] According to an embodiment of the present invention, the importing of the training data into a CNN-based image recognition model for training is specifically as follows:

[0147] Build a CNN-based image recognition model;

[0148] Importing thermal imaging image data in the historical test image data into the image recognition model to perform a model training and achieve a first preset recognition rate;

[0149] The training data is imported into the image recognition model for secondary model training to achieve a second preset recognition rate.

[0150] It should be noted that the second preset recognition rate is greater than the first preset recognition rate.

[0151] According to an embodiment of the present invention, the real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing solution correction information, specifically:

[0152] Acquire real-time test image data of the target circuit board in real time during N consecutive test time periods, wherein the real-time test image data includes real-time high-definition image data and real-time thermal imaging image data;

[0153] The real-time thermal imaging image data is subjected to image noise reduction, smoothing, and standardization preprocessing and imported into an image recognition model for abnormality recognition based on thermal imaging;

[0154] According to the circuit area and the component area, the circuit abnormal area and the component abnormal area are respectively identified;

[0155] According to the real-time analysis of line abnormal areas, component abnormal areas and N continuous test time periods, the number of abnormal areas and abnormal duration in each continuous test time period are analyzed;

[0156] Generate test reports based on the number of abnormal areas, abnormal duration, line abnormal areas and component abnormal areas.

[0157] It should be noted that the test report includes abnormal reports of various areas of the target circuit board in multiple continuous test time periods, which can effectively reflect the temperature abnormality areas after a test and adjust the plan based on the test report. The test time period is generally a functional test with power on.

[0158] In addition, through the regional image model analysis of the present invention, global temperature thermal imaging analysis can be performed on the entire circuit board, and it is not limited to the temperature abnormality monitoring and detection of a certain point. For example, the temperature abnormality of a component affects the temperature of the remaining adjacent areas, causing the temperature of the remaining adjacent areas to fluctuate to a certain extent. Through the present invention, the relevant abnormal temperature control areas can be identified and marked in real time, further realizing the intelligent and automated testing process of the circuit board.

[0159] According to an embodiment of the present invention, the real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing solution correction information, specifically:

[0160] Based on the test report, the circuit area and component area of ​​the target circuit board are processed and evaluated to generate processing evaluation data;

[0161] Based on the processing evaluation data, the processing plan adjustment analysis and circuit board design correction analysis are performed to obtain processing plan correction information.

[0162] It should be noted that the processing evaluation data includes processing evaluation and design rationality evaluation data for circuits and components.

[0163] According to an embodiment of the present invention, it also includes:

[0164] Based on a preset test cycle and in combination with a test plan, M subdivision cycles are divided, and the division is a continuous division based on time;

[0165] The target circuit board is tested based on a preset test cycle, and M pieces of thermal imaging real-time data corresponding to M subdivided cycles are obtained in real time;

[0166] Based on the system database, retrieve and obtain the standard test thermal imaging data within M subdivision periods;

[0167] Importing the standard test thermal imaging data within the M subdivided periods into an image recognition model for feature recognition and extraction to obtain M standard feature data;

[0168] The standard feature data includes color features and thermal imaging contour features of the circuit area and component area;

[0169] Importing the M pieces of thermal imaging real-time data into an image recognition model for feature recognition and extraction to obtain M pieces of real-time feature data;

[0170] The real-time feature data includes color features and thermal imaging contour features of the circuit area and component area;

[0171] Taking each segmentation period as the analysis unit, M real-time feature data and M standard feature data are compared and analyzed, and the similarities of M features are calculated based on the Mahalanobis distance;

[0172] Based on a preset test cycle and M sets of real-time thermal imaging data, the temperature fluctuation data of the target circuit board test process is obtained;

[0173] The temperature fluctuation data includes temperature fluctuation curves of the circuit area and the component area;

[0174] Acquire standard temperature fluctuation data based on standard test thermal imaging data;

[0175] Based on the temperature fluctuation data and the standard temperature fluctuation data, a periodic temperature difference analysis based on M periods is performed to obtain M temperature control difference degrees;

[0176] Based on M feature similarities and M temperature control differences, life and quality assessment of the target circuit board are performed to obtain life assessment information and quality assessment information.

[0177] It should be noted that the preset test cycle is a longer cycle, and refined analysis is performed by dividing each time period. The test plan includes the test duration and test intensity of different components in different time periods within a preset test cycle. The M feature similarities and M temperature control differences can reflect the test differences between the target circuit board and the standard production standard finished product. The higher the feature similarity or the lower the temperature control difference, the better the temperature control of the target circuit board, the higher the quality of the finished product, and the higher the corresponding finished product life. The feature similarity and temperature control difference are respectively used to evaluate the quality of the target circuit board in the regional feature dimension and the temperature curve change dimension, realizing a dual-dimensional circuit board abnormality analysis. It is worth mentioning that through the dual-dimensional analysis, the small temperature control differences in the circuit board can be amplified. The difference analysis can be realized, and the precise and accurate testing process can be further realized.

[0178] The third aspect of the present invention also provides a computer-readable storage medium, which includes a deep learning-based test analysis program. When the deep learning-based test analysis program is executed by a processor, it implements the steps of the deep learning-based test analysis method as described in any one of the above items.

[0179] The present invention discloses a test analysis method, system and storage medium based on deep learning, which obtains historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data; the historical test image data is subjected to image preprocessing and imported into an adversarial neural network model based on deep learning for generating and analyzing training data, and training data is generated in combination with a preset circuit test plan; the training data is imported into a CNN-based image recognition model for training, and real-time test image data of a target circuit board is obtained in real time; the real-time test image data is imported into an image recognition model for abnormal area recognition, and a test report is generated, and the target circuit board is evaluated and analyzed based on the test report to obtain processing evaluation data and processing scheme correction information. The present invention can realize real-time, rapid and accurate detection of circuit board defects, thereby achieving the purpose of rapidly improving the quality of finished products of circuit boards.

[0180] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0181] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0182] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0183] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0184] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0185] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A test analysis method based on deep learning, characterized in that: include: Acquire historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data; The historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan; Importing the training data into a CNN-based image recognition model for training, and acquiring real-time test image data of the target circuit board in real time; Importing the real-time test image data into an image recognition model to identify abnormal areas, and generating a test report, evaluating and analyzing the target circuit board based on the test report, and obtaining processing evaluation data and processing plan correction information; The process of preprocessing the historical test image data and importing the historical test image data into a deep learning-based adversarial neural network model to generate training data and generate training data in combination with a preset circuit test plan includes: Perform image denoising, smoothing and standardization preprocessing on historical test image data; Based on the circuit board component area information, the thermal imaging image data is divided into regions to obtain the circuit area and the component area; Based on the circuit area and the component area, color features are extracted from the thermal imaging image data to obtain circuit color features and component color features; Perform temperature range analysis on the component area based on the thermal imaging image data to obtain the temperature variation range, and divide N temperature span ranges based on the temperature variation range; Based on the N temperature span ranges, extracting contour features of the component area of ​​the thermal imaging image data to obtain N contour feature data, each contour feature data corresponding to a temperature span range; The historical test image data is subjected to image preprocessing and imported into a deep learning-based adversarial neural network model to generate and analyze training data, and training data is generated in combination with a preset circuit test plan, specifically: Build an adversarial neural network model based on deep learning and label it as an adversarial generative model; Building generators and discriminators in adversarial generative models; Import the thermal imaging image data in the historical test image data into the adversarial generation model and use it as real data for adversarial training; Vectorize the line color features, component color features, and N contour feature data to generate feature vector data, and import the feature vector data into the generator as a generator model parameter; Perform cyclic adversarial training based on the generator and the discriminator until the generator and the discriminator reach Nash equilibrium; A preset amount of thermal imaging simulation feature data is generated based on the trained adversarial generative model, and the thermal imaging simulation feature data is used as training data.

2. A test analysis method based on deep learning according to claim 1, characterized in that: The obtaining of historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data, is specifically: Acquire historical test image data of a preset circuit board, and divide the historical test image data into high-definition image data and thermal imaging image data; Based on the design information of the preset circuit board, obtain the circuit board component area information; Based on the circuit board component area information, the high-definition image data and the thermal imaging image data are associated between the image areas.

3. A test analysis method based on deep learning according to claim 1, characterized in that: The training data is imported into the CNN image recognition model for training, specifically: Build a CNN-based image recognition model; Importing thermal imaging image data in the historical test image data into the image recognition model to perform a model training and achieve a first preset recognition rate; The training data is imported into the image recognition model for secondary model training to achieve a second preset recognition rate.

4. A test analysis method based on deep learning according to claim 3, characterized in that: The real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing plan correction information, specifically: Acquire real-time test image data of the target circuit board in real time during N consecutive test time periods, wherein the real-time test image data includes real-time high-definition image data and real-time thermal imaging image data; The real-time thermal imaging image data is subjected to image noise reduction, smoothing, and standardization preprocessing and imported into an image recognition model for abnormality recognition based on thermal imaging; According to the circuit area and the component area, the circuit abnormal area and the component abnormal area are respectively identified; According to the real-time analysis of line abnormal areas, component abnormal areas and N continuous test time periods, the number of abnormal areas and abnormal duration in each continuous test time period are analyzed; Generate test reports based on the number of abnormal areas, abnormal duration, line abnormal areas and component abnormal areas.

5. A test analysis method based on deep learning according to claim 4, characterized in that: The real-time test image data is imported into the image recognition model to identify abnormal areas, and a test report is generated. Based on the test report, the target circuit board is evaluated and analyzed to obtain processing evaluation data and processing plan correction information, specifically: Based on the test report, the circuit area and component area of ​​the target circuit board are processed and evaluated to generate processing evaluation data; Based on the processing evaluation data, the processing plan adjustment analysis and circuit board design correction analysis are performed to obtain processing plan correction information.

6. A test analysis system based on deep learning, characterized in that: The system includes: a memory and a processor, wherein the memory includes a deep learning-based test analysis program, and when the deep learning-based test analysis program is executed by the processor, the steps of the deep learning-based test analysis method as described in claim 1 are implemented.

7. A test analysis system based on deep learning according to claim 6, characterized in that: The obtaining of historical test image data of a preset circuit board, wherein the test image data includes high-definition image data and thermal imaging image data, is specifically: Acquire historical test image data of a preset circuit board, and divide the historical test image data into high-definition image data and thermal imaging image data; Based on the design information of the preset circuit board, obtain the circuit board component area information; Based on the circuit board component area information, the high-definition image data and the thermal imaging image data are associated between the image areas.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a deep learning-based test analysis program. When the deep learning-based test analysis program is executed by a processor, the steps of the deep learning-based test analysis method as described in any one of claims 1 to 5 are implemented.

Images