A crack detection network based on deformable convolution and test time augmentation

Guoliang Zhang; Zexu Du; Peng Wu; Xiangquan Zhang; Weizhou Wang; Zhiyuan Wang

doi:10.1117/12.2640993

2 December 2022 A crack detection network based on deformable convolution and test time augmentation

Guoliang Zhang, Zexu Du, Peng Wu, Xiangquan Zhang, Weizhou Wang, Zhiyuan Wang

Proceedings Volume 12288, International Conference on Computer, Artificial Intelligence, and Control Engineering (CAICE 2022); 122880A (2022) https://doi.org/10.1117/12.2640993
Event: International Conference on Computer, Artificial Intelligence, and Control Engineering (CAICE 2022), 2022, Zhuhai, China

Abstract

This paper proposes an improved crack detection network, which can provide convenience in pavement maintenance and reduce the burden of labor costs. The network architecture is built based on YOLOv5-l, which has a balance of accuracy and speed. We introduce deformable convolution into the designed network to adapt to the crack target, which can improve the effect of detection. During the test, we also use test time augmentation to further improve the detection performance. The results on the RDDC2020 dataset show that the proposed model has reached a high level without using data augmentation and ensemble learning.

Citation Download Citation

Guoliang Zhang, Zexu Du, Peng Wu, Xiangquan Zhang, Weizhou Wang, and Zhiyuan Wang "A crack detection network based on deformable convolution and test time augmentation", Proc. SPIE 12288, International Conference on Computer, Artificial Intelligence, and Control Engineering (CAICE 2022), 122880A (2 December 2022); https://doi.org/10.1117/12.2640993

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