CN-122023298-A - Screen flower screen detection method, device and storage medium

CN122023298ACN 122023298 ACN122023298 ACN 122023298ACN-122023298-A

Abstract

The application relates to a screen pattern detection method, a screen pattern detection device and a storage medium. The method comprises the steps of obtaining screen display picture data, inputting the screen display picture data into a pre-trained self-supervision image reconstruction model combining multi-scale feature pyramid fusion and channel space dual attention, outputting a reconstruction image and a space attention thermodynamic diagram, determining reconstruction errors according to the input image and the reconstruction image, generating an abnormal attention thermodynamic diagram according to the space attention thermodynamic diagram, fusing the reconstruction errors and the abnormal attention thermodynamic diagram, calculating to obtain corresponding image abnormal scores, comparing the image abnormal scores with a preset threshold, and judging whether a screen is stained according to a comparison result. By adopting the method, the cost can be reduced, the environmental robustness of the detection method is improved, and the high-precision identification of the complex screen is realized.

Inventors

LI GUILIN
LI GANG
YAO XIAN
LIANG LUYAO
HE HAO

Assignees

华兴源创(成都)科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (10)

1. A screen splash screen detection method, characterized in that the method comprises: acquiring screen display picture data; inputting the screen display picture data into a pre-trained self-supervision image reconstruction model combining multi-scale feature pyramid fusion and channel space dual attention, and outputting a reconstructed image and a space attention thermodynamic diagram; Determining a reconstruction error according to an input image and a reconstruction image, generating an abnormal attention thermodynamic diagram according to the spatial attention thermodynamic diagram, fusing the reconstruction error and the abnormal attention thermodynamic diagram, and calculating to obtain a corresponding image anomaly score; And comparing the abnormal score of the image with a preset threshold value, and judging whether the screen is stained or not according to a comparison result.
2. The method according to claim 1, wherein the method further comprises: And carrying out any one or more processing modes of normalization processing, noise reduction processing and sliding window cutting processing on the screen display picture data to generate a sub-image suitable for the input of the self-supervision image reconstruction model.
3. The method according to claim 2, wherein the comparing the image anomaly score with a preset threshold value, and determining whether the screen is a screen splash according to the comparison result, comprises: And taking the maximum value of the abnormal scores of all the sub-images, summing the maximum value, determining the overall abnormal index of the screen, comparing the overall abnormal index with a preset threshold value, and judging whether the screen is jumped or not according to the comparison result.
4. The method of claim 1, wherein determining a reconstruction error from the input image and the reconstructed image comprises: Carrying out weighted summation on the first error and the second error to determine a reconstruction error; the first error characterizes a mean square error of the input image and the reconstructed image, and the second error characterizes a similarity error of the input image and the reconstructed image.
5. The method of claim 1, wherein generating an abnormal-attention thermodynamic diagram from the spatial-attention thermodynamic diagram comprises: and taking the output thermodynamic diagram of the spatial attention module in the bottleneck layer of the self-supervision image reconstruction model as a spatial attention thermodynamic diagram, calculating local entropy, and taking the local entropy as an abnormal attention thermodynamic diagram.
6. The method of claim 4, wherein the bottleneck layer is between an encoder and a decoder of the self-supervised image reconstruction model, the bottleneck layer configured to introduce a channel attention mechanism and a spatial attention mechanism.
7. The method of claim 1, wherein the decoder of the self-supervised image reconstruction model is configured to fuse feature maps of different stages of an encoder with upsampling results of the decoder's corresponding layer via a skip connection.
8. The method of claim 1, wherein the fusing the reconstruction error and the abnormal-attention thermodynamic diagram and calculating a corresponding image anomaly score comprises: element multiplication is carried out on the reconstruction error and the abnormal attention thermodynamic diagram, and a final error diagram is obtained; Respectively carrying out global average pooling and global maximum pooling on the final error map to respectively obtain a basic anomaly score and a maximum anomaly score; And carrying out weighted summation on the basic anomaly score and the maximum anomaly score, and calculating to obtain the corresponding image anomaly score.
9. A screen splash screen detection device, the device comprising: The data acquisition module is used for acquiring screen display picture data; the self-supervision image reconstruction module is used for inputting the screen display picture data into a pre-trained self-supervision image reconstruction model combining multi-scale feature pyramid fusion and channel space dual attention, and outputting a reconstructed image and a space attention thermodynamic diagram; The abnormal score output module is used for determining a reconstruction error according to an input image and a reconstruction image, generating an abnormal attention thermodynamic diagram according to the spatial attention thermodynamic diagram, fusing the reconstruction error and the abnormal attention thermodynamic diagram, and calculating a corresponding image abnormal score; And the screen-splash detection module is used for comparing the abnormal score of the image with a preset threshold value and judging whether the screen is splash or not according to the comparison result.
10. A computer-readable storage medium, characterized in that a computer program is stored thereon, which, when being executed by a processor, implements the steps of the method of any of claims 1 to 8.

Description

Screen flower screen detection method, device and storage medium Technical Field The present application relates to the field of screen display detection technologies, and in particular, to a screen display detection method, device and storage medium. Background In the display screen production, demura is a core process, and compensation values are generated to eliminate Mura (luminance/color unevenness) by collecting luminance/chromaticity data of a screen display image. The process relies on a high-definition camera to shoot a screen display picture (such as a multi-gray-scale test chart), and compensation parameters are calculated after original data are acquired. In the traditional technology, the method mainly comprises three modes of manual visual inspection, traditional image inspection and supervised learning inspection, wherein the manual visual inspection relies on workers to directly observe a screen display picture to judge whether a screen is stained, the traditional image inspection analyzes screen images to identify the screen by means of a conventional image identification algorithm, the supervised learning inspection is carried out by constructing a neural network model, training the model by using a marked screen sample, and then detecting whether the screen is stained by using the model. However, the pattern screen detection modes have obvious defects of low manual visual inspection efficiency, strong subjectivity and difficulty in adapting to the real-time detection requirement of a high-speed production line, the traditional image detection is affected by pattern screen shapes in various ways and is sensitive to environmental interference, misjudgment is easy to occur, the production efficiency is affected, the supervision and learning detection depend on a large number of marked pattern screen samples, the pattern screen in the actual production line belongs to occasional defects, the samples are rare and have high marking cost, and meanwhile, the generalization capability of a model on novel pattern screens is weak and the dynamic change of the defect shapes of the production line is difficult to adapt. Disclosure of Invention Based on the above, it is necessary to provide a screen splash screen detection method, device and storage medium capable of reducing cost, improving environmental robustness of the detection method and realizing high-precision recognition of complex splash screens. In a first aspect, the application provides a screen splash screen detection method. The method comprises the following steps: acquiring screen display picture data; inputting the screen display picture data into a pre-trained self-supervision image reconstruction model combining multi-scale feature pyramid fusion and channel space dual attention, and outputting a reconstructed image and a space attention thermodynamic diagram; Determining a reconstruction error according to an input image and a reconstruction image, generating an abnormal attention thermodynamic diagram according to the spatial attention thermodynamic diagram, fusing the reconstruction error and the abnormal attention thermodynamic diagram, and calculating to obtain a corresponding image anomaly score; And comparing the abnormal score of the image with a preset threshold value, and judging whether the screen is stained or not according to a comparison result. In some embodiments of the method, the method further comprises: And carrying out any one or more processing modes of normalization processing, noise reduction processing and sliding window cutting processing on the screen display picture data to generate a sub-image suitable for the input of the self-supervision image reconstruction model. In some embodiments of the method, the comparing the abnormal score of the image with a preset threshold value, and determining whether the screen is stained according to the comparison result includes: And taking the maximum value of the abnormal scores of all the sub-images, summing the maximum value, determining the overall abnormal index of the screen, comparing the overall abnormal index with a preset threshold value, and judging whether the screen is jumped or not according to the comparison result. In some embodiments of the method, the determining a reconstruction error from the input image and the reconstructed image comprises: Carrying out weighted summation on the first error and the second error to determine a reconstruction error; the first error characterizes a mean square error of the input image and the reconstructed image, and the second error characterizes a similarity error of the input image and the reconstructed image. In some embodiments of the method, the generating an abnormal attention thermodynamic diagram from the spatial attention thermodynamic diagram comprises: and taking the output thermodynamic diagram of the spatial attention module in the bottleneck layer of the self-supervision image reconstruction model as