Neighborhood Correlation Enhancement Network for PCB Defect Classification
In recent years, deep-learning has been gradually applied to the detection and classification of true and pseudo defects in printed circuit boards (PCB). To judge the authenticity of PCB defects, it is necessary not only to combine the shape characteristics of the defects themselves, but also to add...
Saved in:
Published in | IEEE transactions on instrumentation and measurement Vol. 72; p. 1 |
---|---|
Main Authors | , , , , |
Format | Journal Article |
Language | English |
Published |
New York
IEEE
01.01.2023
The Institute of Electrical and Electronics Engineers, Inc. (IEEE) |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | In recent years, deep-learning has been gradually applied to the detection and classification of true and pseudo defects in printed circuit boards (PCB). To judge the authenticity of PCB defects, it is necessary not only to combine the shape characteristics of the defects themselves, but also to add the relationship information between the surrounding environment and defects. However, it is difficult for general classification methods to extract such features. In this paper, we propose the neighborhood correlation enhancement network (NCE-Net), which effectively uses defect and surrounding relationship information to accurately distinguish defect authenticity. This network has a relevance residual block (RRB), which is used to establish correlation between defects and their surroundings, including the location enhancement and locate module (LEL) and relevance convolution (RC), which are respectively used to enhance the effective geographical information and extract the relationship between features of different positions. It also utilizes a small squeeze residual block (SRB) to classify pseudo defects more quickly and efficiently in industrial applications. In addition, to achieve the lowest pseudo defect detection error rate, we created a unique multi-network specific integration (SI) method for use with NCE-Net. The experimental results show that our proposed network can be trained on a PCB defect classification dataset (PCB-2-DET) for higher efficiency and more significant PCB defect detection. Additionally, the model's identification accuracy can be further improved through our unique SI method. |
---|---|
ISSN: | 0018-9456 1557-9662 |
DOI: | 10.1109/TIM.2023.3243670 |