Image recognition method and system with improved accuracy of recognizing defect, and training method and an electronic device

A training method for an image recognition model of a display panel, comprising:

• obtaining an image of the display panel and annotation data of the image, wherein the image of the display panel comprises a plurality of gate lines and a plurality of data lines, the plurality of gate lines extending in a first direction, the plurality of data lines extending in a second direction different from the first direction, and the plurality of gate lines and the plurality of data lines intersecting to define a plurality of sub-pixel regions, the image of the display panel further comprises a defect pattern, the defect pattern being in one or more of the plurality of sub-pixel regions, and the annotation data comprises at least one of an actual defect type or an actual defect position of the defect pattern;
• recognizing the defect pattern in the image of the display panel by using the image recognition model to obtain defect information, comprising:
• performing dimensionality reduction processing on the image of the display panel to obtain a pixel encoding of the image of the display panel, the pixel encoding representing a pixel feature of the image of the display panel;
• determining a position encoding of the image of the display panel, the position encoding representing a positional relationship feature between different pixels in the image of the display panel;
• determining a feature encoding of the image of the display panel based on the pixel encoding and the position encoding, the feature encoding integrating the pixel feature of the image of the display panel and the positional relationship feature between the different pixels;
• processing the feature encoding by using a first attention model to obtain an output of the first attention model, the first attention model being configured to learn a weight proportion of the feature of the defect pattern in the image of the display panel based on the pixel feature of the image of the display panel and the positional relationship feature between the different pixels;
• and
• determining the defect information based on the output of the first attention model, wherein the defect information comprises at least one of a defect type or a defect position of the defect pattern, wherein the image recognition model comprises the first attention model;
• and
• training the image recognition model based on the defect information and the annotation data to obtain a trained image recognition model.

An image recognition system for a display panel, comprising:

• a server configured to:
• obtain an image of the display panel, wherein the image of the display panel comprises a plurality of gate lines and a plurality of data lines, the plurality of gate lines extending in a first direction, the plurality of data lines extending in a second direction different from the first direction, and the plurality of gate lines and the plurality of data lines intersecting to define a plurality of sub-pixel regions, and the image of the display panel further comprises a defect pattern, the defect pattern being in one or more of the plurality of sub-pixel regions;
• and
• recognize the defect pattern in the image of the display panel by using an image recognition model to obtain defect information,
• wherein the server is configured to:
• perform dimensionality reduction processing on the image of the display panel to obtain a pixel encoding of the image of the display panel, the pixel encoding representing a pixel feature of the image of the display panel;
• determine a position encoding of the image of the display panel, the position encoding representing a positional relationship feature between different pixels in the image of the display panel;
• determine a feature encoding of the image of the display panel based on the pixel encoding and the position encoding, the feature encoding integrating the pixel feature of the image of the display panel and the positional relationship feature between the different pixels;
• process the feature encoding by using a first attention model to obtain an output of the first attention model, the first attention model being configured to learn a weight proportion of the feature of the defect pattern in the image of the display panel based on the pixel feature of the image of the display panel and the positional relationship feature between the different pixels;
• and
• determine the defect information based on the output of the first attention model,
• wherein the defect information comprises at least one of a defect type or a defect position of the defect pattern, wherein the image recognition model comprises the first attention model.