Thursday, May 2025

05:00 PM - 08:00 PM

Room: 220A

Session: Artificial Intelligence Including Machine Learning for Imaging Posters

Exploration and Application of Unknown Category Defect Detection Methods for Display Panels

Description:

Defect detection is crucial in the panel industry, especially for unknown defects that occur during the production process. In order to solve this problem, we propose a novel feature fusion method, the Foreground-background Same-scale Difference Attention (FBSSDA) fusion module. Improvements are made to YOLOv5, and the FBSSDA-YOLOv5 defect detection model is proposed to be applied to the production inspection of display panels. First, in the feature extraction stage, the model constructs a feature fusion module by introducing the FBSSDA attention mechanism to extract foreground and background difference features at the same scale; second, a contrast loss function is introduced to expand the differences between different defect categories; finally, an improved NMS algorithm, U-NMS, is proposed to increase the recognition of unknown defects.