Premium
41.4: Display Panel Defect Detection Algorithm Based On Group Convolutions
Author(s) -
Ma WeiFei,
Zhang ShengSen,
Zheng ZengQiang
Publication year - 2019
Publication title -
sid symposium digest of technical papers
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.351
H-Index - 44
eISSN - 2168-0159
pISSN - 0097-966X
DOI - 10.1002/sdtp.13530
Subject(s) - computer science , convolution (computer science) , process (computing) , algorithm , network packet , feature (linguistics) , artificial intelligence , flat panel display , feature extraction , pattern recognition (psychology) , artificial neural network , computer network , linguistics , philosophy , operating system
The traditional image processing algorithm can not detect the defects in the display panel well, but the application scenarios of the mainstream deep learning detection model are mostly natural scenes, and can not be directly applied to the detection of display panel defects. Aiming at this problem, this paper proposes a display panel defect detection algorithm based on packet convolution network. The algorithm uses YOLO3 network as the basic architecture. It is aimed at the lack of training samples and the large resolution of single‐defect images in the process of defect detection in industrial display panels. By referring to the dense connection and packet convolution characteristics in the CondenseNet network model, YOLO3's backbone feature extraction network has been improved to reduce the need for the number of training samples for the test model and to speed up the prediction of a single image. Finally, through a series of improvements to the YOLO3 network, this paper proposes a special name called DPDDNet. A network model for displaying panel defect detection. The experimental results show that compared with the traditional algorithm, the proposed algorithm has fewer configuration parameters and faster prediction speed, and is more suitable for defect detection of display panels.