Automatic detection method for abnormal display of display screen in ESD test
By leveraging the collaborative work of template image extraction, adaptive threshold analysis, and anomaly detection modules, automated detection of abnormal displays during ESD testing is achieved. This solves the problems of poor repeatability, low efficiency, and strong subjectivity in traditional manual testing, thereby improving detection accuracy and efficiency.
Patent Information
- Application Number
- CN202511190494.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-25
- Publication Date
- 2025-12-09
AI Technical Summary
Traditional manual ESD testing methods suffer from poor repeatability, low efficiency, strong subjectivity, and poor traceability. They cannot achieve automated detection of abnormal display conditions, especially dynamic display abnormalities such as screen flickering, dimming, and random display. The detection efficiency is low and the standards are not uniform.
The system employs a template image extraction module, an adaptive threshold analysis module, and an anomaly detection module. It acquires images using a CCD industrial camera, analyzes ROI region features based on the RGB color model, and utilizes an autoencoder and DBSCAN density clustering algorithm to achieve automated detection, generating adaptive thresholds and discrimination results.
It has achieved automated and accurate detection of abnormal display screens, improved detection efficiency and accuracy, reduced manual intervention and subjective errors, and built a fully automated detection system for ESD testing.
Smart Images
Figure SMS_1 
Figure SMS_6 
Figure SMS_8
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of machine vision and image processing, and particularly relates to an automatic detection method for abnormal display of a display screen in ESD testing. BACKGROUND
[0002] Electrostatic discharge (ESD) refers to an event in which a limited amount of electric charge is transferred from one object to another object, and as a common near-field hazard source, ESD not only directly or indirectly interferes with semiconductor devices or electronic systems to cause damage, but also forms potential hazards to the reliability thereof.
[0003] The traditional manual ESD testing method has the following problems: the repeatability of the dotting action is high, the operator is prone to fatigue, and the repeatability of the test result is poor; there are test differences between operators, and there is subjectivity in product abnormal grade judgment and quality evaluation; traceability is poor, and it is not possible to accurately trace the details and results of each test; manual operation is inefficient, which affects production efficiency and cost control.
[0004] In the prior art, there is no mature solution for the detection method of dynamic display abnormalities such as flashing screen, dimming, disorderly display, and screen flower during ESD testing, and manual judgment is mainly relied on, which has problems such as non-uniform judgment standard, low detection efficiency, and inability to realize automatic testing. Therefore, it is urgent to develop a detection method capable of automatically identifying abnormal display of a display screen in ESD testing to improve testing efficiency and detection accuracy. SUMMARY
[0005] The main purpose of the present application is to overcome the above-mentioned defects in the prior art, and to provide an automatic detection method for abnormal display of a display screen in ESD testing, which solves the problems of poor repeatability, low efficiency, strong subjectivity, and poor traceability of the traditional manual testing method, and realizes automatic and accurate detection of display abnormalities during ESD testing.
[0006] In order to achieve the above purpose, the technical scheme of the present application is as follows:
[0007] An automatic detection method for abnormal display of a display screen in ESD test, based on a template image extraction module, an adaptive threshold analysis module and an abnormality discrimination module. Among them, the template image extraction module is responsible for automatically extracting the original image template from the display screen carousel picture, collecting the image sequence of the display screen under normal working state through the CCD industrial camera, analyzing the ROI region characteristics based on the RGB color model, intelligently identifying the carousel period and equally dividing, and selecting the representative picture of each interval as the standard template; the adaptive threshold analysis module first determines the image category by using the difference image method, then compresses the difference image by using the autoencoder, removes noise and redundant information, and finally automatically determines the optimal judgment threshold of each type of image by using the DBSCAN density clustering algorithm; the abnormality discrimination module calculates the similarity index R' α and R' β of the test image and the template, determines the category of the image, compares the similarity index with the corresponding adaptive threshold, and finally outputs the abnormality discrimination result.
[0008] The specific detection method comprises the following steps:
[0009] 1) Collecting the original image data of the carousel of the display screen in the ESD test process and the picture data of the historical test;
[0010] 2) Using the template image extraction module, extracting a round of picture data by analyzing the RGB mean value of the ROI region, and equally dividing it, and selecting the original image template of each interval;
[0011] 3) Calculating the difference value of the test image and each original image template by using the difference image method, and determining the category of the test image according to the minimum difference value;
[0012] 4) Inputting the difference image into the autoencoder for dimension reduction processing, and calculating the similarity index R' α and R' β after dimension reduction;
[0013] 5) Based on the picture data of the historical test, using the adaptive threshold analysis module to normalize the similarity index of the same type of image, using the DBSCAN density clustering algorithm to determine the judgment threshold of each type of image, and generating a threshold database;
[0014] 6) Inputting the test image into the abnormality discrimination module, and comparing the similarity index of the test image with the threshold value of the corresponding category to complete the judgment of abnormal display.
[0015] Preferably, step 2) specifically comprises:
[0016] 2.1) Analyzing the RGB mean value of the first original image in the ROI region to determine the limited range of the RGB mean value;
[0017] 2.2) When the ROI region RGB mean of k pictures is continuously read and meets the defined condition, save the k pictures and subsequent pictures as a round of original picture data;
[0018] 2.3) Equidistantly segment the original picture data to divide it into N picture intervals;
[0019] 2.4) Select the picture in the middle of the interval order as the original picture template.
[0020] Preferably, k and N are positive integers, and k > N.
[0021] Preferably, step 3) specifically includes:
[0022] 3.1) Calculate the difference image of the test image I and the nth template image I n , n = 1, 2, 3,..., N;
[0023] 3.2) Sum all pixel values in the difference image of the three channels respectively;
[0024] 3.3) Calculate the total pixel sum of the three channels;
[0025] 3.4) Select the index with the smallest sum value as the image category.
[0026] Preferably, the calculation method of the similarity index in step 4) is as follows:
[0027] 4.1) Calculate the correlation coefficients R α and R β
[0028]
[0029] Where (x', y') is the coordinate in the template image, (x, y) is the coordinate in the test image, T(x', y') represents the pixel value at a specific position (x', y') in the template image; I(x, y) represents the pixel value at a specific position (x, y) in the test image; w and h are the dimensions of the image;
[0030] 4.2) Normalize the correlation coefficients to scale all values to the range [0, 1] to obtain R' α and R' β .
[0031] Preferably, step 5) specifically includes:
[0032] 5.1) Input the historical difference image into the autoencoder, map the difference image to a low-dimensional representation space through the encoder, and reconstruct the data through the decoder to remove noise and redundant information;
[0033] 5.2) Calculate the similarity index R′ between the compressed difference image and the corresponding template image. α and R′ β ;
[0034] 5.3) Normalize the similarity index of images of the same type;
[0035] 5.4) Use the DBSCAN algorithm to perform density clustering on the normalized data;
[0036] 5.5) Perform inverse normalization on the clustering results to obtain the judgment threshold for this type of image.
[0037] Preferably, the normalization process is a min-max normalization process.
[0038] Preferably, the neighborhood radius eps of the density clustering ranges from 0.01 to 0.1, and the minimum number of neighborhood points MinPts ranges from 3 to 10.
[0039] Preferably, step 6) specifically includes the following steps:
[0040] 6.1) Based on the image category determined in step 3), obtain the template category corresponding to the image to be tested;
[0041] 6.2) Extract the similarity index R′ of the image to be tested using the data obtained in step 4). α and R′ β ;
[0042] 6.3) Extract the threshold corresponding to the category from the threshold database generated in step 5). α and Threshold β ;
[0043] 6.4) Compare the similarity index of the image to be tested with the threshold of the corresponding category and output the discrimination result.
[0044] Preferably, if R′ α ≥Threshold α And R′ β ≥Threshold β If R′ is normal, it is considered a normal display; if R′ is normal, it is considered a normal display. α <Threshold α or R′ β <Threshold β If so, it is determined to be an abnormal display.
[0045] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:
[0046] The automatic detection method for abnormal display of a display screen in ESD testing provided by the application realizes automatic generation of an intelligent original template based on an RGB color model through a template image extraction module, avoids manual intervention and subjective errors, realizes high-precision feature extraction and adaptive threshold determination based on self-encoding clustering through an adaptive threshold analysis module, realizes rapid and accurate detection of real-time images through an abnormality discrimination module, and can effectively distinguish between normal display and abnormal display. The three core modules work together to build a complete ESD testing automatic detection system, realize full-process automation from template establishment to abnormality discrimination, effectively solve the problems of poor repeatability, low efficiency, strong subjectivity and poor traceability of traditional ESD testing methods, and are suitable for large-scale industrial application of display screen ESD immunity testing.
[0047] Other features and advantages of the present application will be described in the following specification, and some will become apparent from the specification, or will be understood through implementation of the present application. BRIEF DESCRIPTION OF DRAWINGS
[0048] Figure 1 The automatic detection method for abnormal display of a display screen in ESD testing provided by the application is a system architecture diagram, which details the working process and data flow of the three core modules of the template image extraction module, the adaptive threshold analysis module and the abnormality discrimination module, and highlights the core role of the abnormality detection algorithm and the visual output of the abnormality discrimination. DETAILED DESCRIPTION
[0049] The application will be further described through specific embodiments.
[0050] Reference Figure 1 The automatic detection method for abnormal display of a display screen in ESD testing provided by the application is based on a template image extraction module, an adaptive threshold analysis module and an abnormality discrimination module, and the detailed steps are as follows:
[0051] 1) Use the ESD automatic testing platform to complete the collection of picture data, collect the image sequence of the display screen in the normal working state through a CCD industrial camera, and save it in a suitable format for subsequent calling, which serves as the original data and historical test picture data.
[0052] 2) Automatically extract the original template based on the RGB color model, including:
[0053] 2.1) RGB mean analysis
[0054] Analyze the RGB mean of the first original image in the ROI region, and determine the limited range of the RGB mean as Wherein are the red, green and blue channel mean values of the initial image, respectively.
[0055] 2.2) Carousel cycle identification
[0056] When the RGB mean values of the ROI regions of more than k consecutive images meet the specified conditions, these k images and subsequent images are saved as a round of original image data to achieve intelligent carousel cycle recognition.
[0057] 2.3) Equidistant Segmentation
[0058] Suppose we have N template images (for example, N=19). We divide the original image data into N image intervals at equal intervals, ensuring that each interval contains the same number of images.
[0059] 2.4) Template Selection
[0060] Images from the middle of each interval are selected as the original image templates to avoid the influence of images during the transition phase when switching screens in the ESD test, thus ensuring the stability and representativeness of the templates.
[0061] 3) The difference between the image to be tested and the original image template is calculated using the difference method to determine the image category. This includes:
[0062] 3.1) Difference Image Calculation
[0063] Establish an ESD test image difference calculation model to calculate the difference between the test image I and the nth template image I. n The difference image: D n (i, j, k) = |I n (i,j,k)-I(i,j,k)|, where n=1,2,3,...,N, i∈[1,H], j∈[1,W], k∈{R,G,B};
[0064] 3.2) Differential images D of the three channels respectively n Sum all pixel values in:
[0065] 3.3) Calculate the total pixel sum of the three channels:
[0066] 3.4) Select the index with the smallest sum as the image category: Index = argmin{TotalSum1, TotalSum2, ..., TotalSum...} n};
[0067] 4) Calculate the similarity index using a deep autoencoder network model, as detailed below:
[0068] 4.1) Construction of Autoencoder Network
[0069] A deep autoencoder network model suitable for ESD test images is constructed, including an encoder and a decoder, for image feature extraction and dimension reduction.
[0070] 4.2) Correlation coefficient R α and R β Calculation:
[0071]
[0072] where (x', y') is the coordinate in the template image, (x, y) is the coordinate in the test image, T(x', y') represents the pixel value at a specific position (x', y') in the template image; I(x, y) represents the pixel value at a specific position (x, y) in the test image; w and h are the dimensions of the image.
[0073] The correlation coefficient is normalized to scale all values to the range [0, 1], and R' is obtained α and R' is obtained. β .
[0074] 5) Determine the decision threshold using the adaptive threshold analysis module, as follows:
[0075] 5.1) Image dimensionality reduction
[0076] Input the difference image in the ESD test into the pre-trained autoencoder, map the difference image to a 128-dimensional low-dimensional representation space through the encoder, and reconstruct the data through the decoder to remove noise and redundant information generated during the ESD test.
[0077] 5.2) Similarity index calculation
[0078] Calculate the similarity index R' of the compressed difference image and the corresponding template image α and R' is obtained. β .
[0079] 5.3) Data normalization
[0080] Perform min-max normalization on the similarity index of the same type of ESD test image:
[0081]
[0082] 5.4) Density clustering
[0083] Use the DBSCAN algorithm to perform density clustering on the normalized data, set the neighborhood radius eps = 0.1, and the minimum number of neighborhood points MinPts = 5, to automatically identify the data distribution pattern.
[0084] 5.5) Threshold determination
[0085] The clustering result is de-normalized to obtain the determination threshold of the ESD test image of the class, and adaptive threshold optimization is realized. A threshold database containing different categories of images is generated.
[0086] 6) The similarity index of the image to be tested is compared with the threshold of the corresponding category, and the specific implementation is as follows:
[0087] 6.1) Image category acquisition
[0088] According to the image category Index determined in step 3), the template category corresponding to the image to be tested is obtained;
[0089] 6.2) Similarity index acquisition
[0090] The similarity index R' of the image to be tested calculated in step 4) is used α and R' β ;
[0091] 6.3) Threshold data extraction
[0092] The determination threshold Threshold corresponding to the Image category Index is extracted from the threshold database generated in step 5) α and Threshold β ;
[0093] 6.4) Double threshold comparison
[0094] If R' α ≥ Threshold α and R' β ≥ Threshold β , it is determined that the display is normal; if R' α < Threshold α or R' β < Threshold β , it is determined that the display is abnormal.
[0095] Through the above automatic detection method, the template image can be quickly extracted in the ESD test, and the determination threshold of each type of template image can be adaptively analyzed, so that the abnormal display of the display screen can be accurately recognized, which has important application value for guiding the subsequent display screen quality control work.
[0096] The above examples are only used to further illustrate the automatic detection method of the display screen abnormal display in the ESD test of the present application, but the present application is not limited to the examples, and any simple modification, equivalent change and modification according to the technical essence of the present application to the above examples all fall within the protection scope of the technical scheme of the present application.
Claims
1. An automatic detection method for abnormal display on a screen during ESD testing, characterized in that, Based on the template graph extraction module, adaptive threshold analysis module, and anomaly detection module, the automatic detection method includes the following steps: 1) Collect the original image data and historical test image data that are displayed on the screen during the ESD test; 2) Using the template image extraction module, a round of image data is extracted by analyzing the RGB mean of the ROI region, and then the data is divided into equal intervals to select the original image template for each interval; 3) Calculate the difference between the test image and each original image template using the difference method, and determine the category to which the test image belongs based on the minimum difference value; 4) Input the difference image into the autoencoder for dimensionality reduction, and calculate the similarity index R′ after dimensionality reduction. α and R′ β ; 5) Based on historical test image data, the similarity index of similar images is normalized using the adaptive threshold analysis module, and the judgment threshold of each type of image is determined by the DBSCAN density clustering algorithm to generate a threshold database. 6) Input the image to be tested into the anomaly detection module, and compare the similarity index of the image to be tested with the threshold of the corresponding category to complete the determination of anomaly display.
2. The automatic detection method according to claim 1, characterized in that: Step 2) specifically includes: 2.1) Analyze the RGB mean values in the ROI region of the first original image to determine the range of the RGB mean values; 2.2) When the RGB mean values of the ROI regions of k consecutively read images meet the specified conditions, these k images and subsequent images are saved as a round of original image data; 2.3) Divide the original image data into N image intervals using equal intervals; 2.4) Select the image with the middle position in each interval as the original image template.
3. The automatic detection method according to claim 2, characterized in that: k and N are positive integers, and k > N.
4. The automatic detection method according to claim 2, characterized in that, Step 3) specifically includes: 3.1) Calculate the relationship between the test image I and the nth template image I. n The difference image, n = 1, 2, 3, ..., N; 3.2) Sum all pixel values in the difference images of the three channels respectively; 3.3) Calculate the total pixel sum of the three channels; 3.4) Select the index with the smallest sum value as the image category.
5. The automatic detection method according to claim 1, characterized in that, The similarity index in step 4) is calculated as follows: 4.1) Calculate the correlation coefficient R. α and R β Where (x′, y′) are the coordinates in the template image, and (x, y) are the coordinates in the test image. T(x′, y′) represents the pixel value at a specific location (x′, y′) in the template image; I(x,y) represents the pixel value at a specific location (x,y) in the test image; w and h are the dimensions of the image; 4.2) Normalize the correlation coefficients by scaling all values to the range [0, 1] to obtain R′. α and R′ β .
6. The automatic detection method according to claim 1, characterized in that, Step 5) specifically includes: 5.1) Input the historical difference image into the autoencoder, map the difference image to a low-dimensional representation space through the encoder, and reconstruct the data through the decoder to remove noise and redundant information; 5.2) Calculate the similarity index R′ between the compressed difference image and the corresponding template image. α and R′ β ; 5.3) Normalize the similarity index of images of the same type; 5.4) Use the DBSCAN algorithm to perform density clustering on the normalized data; 5.5) Perform inverse normalization on the clustering results to obtain the judgment threshold for this type of image.
7. The automatic detection method according to claim 6, characterized in that: The normalization process is a min-max normalization process.
8. The automatic detection method according to claim 6, characterized in that: The neighborhood radius eps for density clustering ranges from 0.01 to 0.1, and the minimum number of neighborhood points MinPts ranges from 3 to 10.
9. The automatic detection method according to claim 1, characterized in that, Step 6) includes the following steps: 6.1) Based on the image category determined in step 3), obtain the template category corresponding to the image to be tested; 6.2) Extract the similarity index R′ of the image to be tested using the data obtained in step 4). α and R′ β ; 6.3) Extract the threshold corresponding to the category from the threshold database generated in step 5). α and Threshold β ; 6.4) Compare the similarity index of the image to be tested with the threshold of the corresponding category and output the discrimination result.
10. The automatic detection method according to claim 9, characterized in that: If R′ α ≥Threshold α And R′ β ≥Threshold β If R′ is normal, it is considered a normal display; if R′ is normal, it is considered a normal display. α Threshold α or R′ β Threshold β If so, it is determined to be an abnormal display.