Defect detection network based on feature fusion and parallel cascade design

Wenbo Hu; Jue Wang; Bo Li; Hongsheng Deng

doi:10.1117/12.2652308

10 November 2022 Defect detection network based on feature fusion and parallel cascade design

Wenbo Hu, Jue Wang, Bo Li, Hongsheng Deng

Proceedings Volume 12331, International Conference on Mechanisms and Robotics (ICMAR 2022); 1233133 (2022) https://doi.org/10.1117/12.2652308
Event: International Conference on Mechanisms and Robotics (ICMAR 2022), 2022, Zhuhai, China

Abstract

At present, the defect detection link is indispensable in industrial production, and for the problem of low accuracy of existing neural networks in steel surface defect detection, a Faster-RCNN-based steel surface defect detection network (P-RCNN) is proposed in this paper. First, a new backbone network (Resnet-P2) is formed by adding feature fusion and attention mechanism to the original Faster-RCNN backbone network Resnet50, which can make the features transfer more effectively in the backbone network; second, a parallel cascade network is introduced, which aims to train a high-quality detector without reducing the number of high-quality samples. The purpose is to train a high quality detector without reducing the number of high quality samples, which makes the detection accuracy better. The experimental results show that the P-RCNN algorithm is used for steel surface defect detection with better detection performance and the accuracy is improved by 7.6% to 83.4% compared with Faster-RCNN, which can be used as a reference for steel surface defect identification applications.

Citation Download Citation

Wenbo Hu, Jue Wang, Bo Li, and Hongsheng Deng "Defect detection network based on feature fusion and parallel cascade design", Proc. SPIE 12331, International Conference on Mechanisms and Robotics (ICMAR 2022), 1233133 (10 November 2022); https://doi.org/10.1117/12.2652308

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available