Granted Patent B2

US 12,505,533 · App. 18/014,838 · Granted Dec 23, 2025

Inventors: Qiaoke Zhou (Beijing, CN); Bo Li (Beijing, CN); Jing Wang (Beijing, CN); Xiang Li (Beijing, CN); Guojian Qu (Beijing, CN); Bochang Wang (Beijing, CN); Tanhong Zhao (Beijing, CN); Hui Guo (Beijing, CN)

Assignees: BEIJING BOE DISPLAY TECHNOLOGY CO., LTD; BEIJING BOE TECHNOLOGY DEVELOPMENT CO., LTD.

G06T7/001G06T5/70G06V10/20G06V10/60G06V10/764G06V10/82G06T2207/20084G06T2207/30108

Abstract

Claims (61)

1 . A display defect detection method, comprising:

collecting at least one display image of at least one display to be detected;

extracting a plurality of band-limited intrinsic mode function components from a display image in the at least one display image by using a complex variational mode decomposition method;

extracting and fusing, by using a convolutional neural network, the plurality of band-limited intrinsic mode function components extracted from the display image, so as to obtain an average brightness value and a brightness uniformity of the display image; and

determining whether a display to be detected in the at least one display to be detected corresponding to the display image is qualified according to a preset classification rule and the average brightness value and the brightness uniformity of the display image.

2 . The display defect detection method according to claim 1 , wherein the preset classification rule includes:

if the average brightness value of the display image is greater than or equal to a preset brightness value, and the brightness uniformity of the display image is greater than or equal to a preset brightness uniformity, determining that the display to be detected corresponding to the display image is qualified; and

if not, determining that the display to be detected corresponding to the display image is unqualified.

3 . The display defect detection method according to claim 2 , wherein before collecting the at least one display image of the at least one display to be detected, the display defect detection method further comprises:

collecting display images of a plurality of sample displays as sample images;

obtaining reference results of whether the sample displays corresponding to respective sample images are qualified;

determining whether a sample display in the plurality of sample displays that each sample image corresponds to is qualified by using the convolutional neural network, so as to obtain actual detection results of whether the plurality of sample displays are qualified; the convolutional neural network including detection parameters;

comparing an actual detection result of a sample display in the plurality of sample displays with a respective reference result to determine whether the actual detection result is consistent with the respective reference result;

adjusting the detection parameters of the convolutional neural network according to a comparison result; and

adjusting the detection parameters of the convolutional neural network repeatedly, until the actual detection results of the plurality of sample displays are stable.

4 . The display defect detection method according to claim 1 , wherein before collecting the at least one display image of the at least one display to be detected, the display defect detection method further comprises:

collecting display images of a plurality of sample displays as sample images;

obtaining reference results of whether the sample displays corresponding to respective sample images are qualified;

determining whether a sample display in the plurality of sample displays that each sample image corresponds to is qualified by using the convolutional neural network, so as to obtain actual detection results of whether the plurality of sample displays are qualified; the convolutional neural network including detection parameters;

comparing an actual detection result of a sample display in the plurality of sample displays with a respective reference result to determine whether the actual detection result is consistent with the respective reference result;

adjusting the detection parameters of the convolutional neural network according to a comparison result; and

adjusting the detection parameters of the convolutional neural network repeatedly, until the actual detection results of the plurality of sample displays are stable.

5 . The display defect detection method according to claim 4 , wherein before determining whether the sample display that each sample image corresponds to is qualified by using the convolutional neural network, the display defect detection method further comprises:

extracting a plurality of band-limited intrinsic mode function components from the sample image by using the complex variational mode decomposition method.

6 . The display defect detection method according to claim 5 , wherein determining whether the sample display that each sample image corresponds to is qualified by using the convolutional neural network, includes:

extracting and fusing, by using the convolutional neural network, the plurality of band-limited intrinsic mode function components extracted from the sample image, so as to obtain an average brightness value and a brightness uniformity of the sample image; and

determining whether the sample display corresponding to the sample image is qualified according to the preset classification rule and the average brightness value and the brightness uniformity of the sample image.

7 . The display defect detection method according to claim 4 , wherein comparing the actual detection result of the sample display with the respective reference result to determine whether the actual detection result is consistent with the respective reference result, and adjusting the detection parameters of the convolutional neural network according to the comparison result, include:

if the actual detection result of the sample display is inconsistent with the respective reference result, adjusting the detection parameters of the convolutional neural network.

8 . The display defect detection method according to claim 4 , wherein the detection parameters include a kernel function parameter and a penalty parameter.

9 . The display defect detection method according to claim 1 , wherein after collecting the at least one display image of the at least one display to be detected, the display defect detection method further comprises:

preprocessing the display image; the preprocessing including at least one of image cropping, graying and filtering.

10 . The display defect detection method according to claim 1 , wherein after collecting the at least one display image of the at least one display to be detected, the display defect detection method further comprises:

disposing a plurality of monitoring points on the display image; and

obtaining brightness information of the plurality of monitoring points;

wherein the plurality of monitoring points are arranged in an array; a distance between two adjacent monitoring points in a first direction is substantially equal to a distance between two adjacent monitoring points in a second direction; and the first direction and the second direction intersect.

11 . The display defect detection method according to claim 10 , wherein extracting the plurality of band-limited intrinsic mode function components from the display image by using the complex variational mode decomposition method, includes:

decomposing brightness information of each monitoring point on the display image into a plurality of modal components by using the complex variational mode decomposition method; and

removing at least one noise component in the plurality of modal components of the monitoring point to extract band-limited intrinsic mode function components in the plurality of modal components of the monitoring point.

12 . The display defect detection method according to claim 10 , wherein extracting and fusing, by using a convolutional neural network, the plurality of band-limited intrinsic mode function components extracted from the display image to obtain the average brightness value and the brightness uniformity of the display image, includes:

extracting band-limited intrinsic mode function components of brightness information of a monitoring point in the plurality of monitoring points on the display image by using the convolutional neural network, so that a brightness corresponding to the band-limited intrinsic mode function components of the monitoring point is used as the average brightness value of the display image.

13 . The display defect detection method according to claim 10 , wherein extracting and fusing, by using a convolutional neural network, the plurality of band-limited intrinsic mode function components extracted from the display image to obtain the average brightness value and the brightness uniformity of the display image, includes:

extracting band-limited intrinsic mode function components of a monitoring point with least brightness information in the plurality of monitoring points and band-limited intrinsic mode function components of a monitoring point with most brightness information in the plurality of monitoring points; and

calculating a ratio of a brightness corresponding to the band-limited intrinsic mode function components of the monitoring point with least brightness information to a brightness corresponding to the band-limited intrinsic mode function components of the monitoring point with most brightness information, so as to obtain the brightness uniformity of the display image.

14 . The display defect detection method according to claim 10 , wherein extracting and fusing, by using a convolutional neural network, the plurality of band-limited intrinsic mode function components extracted from the display image to obtain the average brightness value and the brightness uniformity of the display image, includes:

extracting band-limited intrinsic mode function components of brightness information of at least two monitoring points in the plurality of monitoring points on the display image by using the convolutional neural network; and

calculating an average value of brightnesses corresponding to the band-limited intrinsic mode function components of the brightness information of the at least two monitoring points as the average brightness value of the display image.

15 . The display defect detection method according to claim 10 , wherein the display image is in a shape of a polygon; extracting and fusing, by using a convolutional neural network, the plurality of band-limited intrinsic mode function components extracted from the display image to obtain the average brightness value and the brightness uniformity of the display image, includes:

extracting band-limited intrinsic mode function components of brightness information of monitoring points including, located at each corner of the display image, at least one corresponding monitoring point in the plurality of monitoring points, and band-limited intrinsic mode function components of brightness information of a monitoring point in the plurality of monitoring points located at a center of the display image by using the convolutional neural network; and

calculating an average value of brightnesses corresponding to the extracted band-limited intrinsic mode function components of the brightness information of the monitoring points including, located at each corner, the at least one corresponding monitoring point, and the monitoring point located at the center as the average brightness value of the display image.

16 . An electronic device comprising a memory and a processor, wherein the memory stores computer program instructions; and

the processor is configured to run the computer program instructions to execute the display defect detection method according to claim 1 .

17 . A non-transitory computer-readable storage medium storing computer program instructions, wherein when the computer program instructions run on a processor, the processor executes the display defect detection method according to claim 1 .

18 . A display defect detection apparatus, comprising:

a feature extractor coupled to an image collector; wherein the feature extractor is configured to: obtain at least one display image of at least one display to be detected collected by the image collector; and extract a plurality of band-limited intrinsic mode function components from a display image in the at least one display image by using a complex variational mode decomposition method; and

a convolutional neural network classifier coupled to the feature extractor; wherein the convolutional neural network classifier is configured to: extract and fuse the plurality of band-limited intrinsic mode function components extracted from the display image, so as to obtain an average brightness value and a brightness uniformity of the display image; and determine whether a display to be detected in the at least one display to be detected corresponding to the display image is qualified according to a preset classification rule and the average brightness value and the brightness uniformity of the display image.

19 . The display defect detection apparatus according to claim 18 , further comprising:

a preprocessor coupled between the image collector and the feature extractor; wherein the preprocessor is configured to: obtain the display image of the display to be detected collected by the image collector; preprocess the display image; and transmit the preprocessed display image to the feature extractor; wherein the preprocessing includes at least one of image cropping, graying and filtering.

20 . A display defect detection system, comprising:

an image collector configured to collect at least one display image of at least one display to be detected; and

the display defect detection apparatus according to claim 18 , the display defect detection apparatus being coupled to the image collector.

Assignments (2)

Priority Claims (1)

Continuity (1)

References Cited (29)