Domain adaptive foggy object detection method based on YOLOv5

Fan Zhang

doi:10.1117/12.2661003

16 December 2022 Domain adaptive foggy object detection method based on YOLOv5

Fan Zhang

Proceedings Volume 12500, Fifth International Conference on Mechatronics and Computer Technology Engineering (MCTE 2022); 125006H (2022) https://doi.org/10.1117/12.2661003
Event: 5th International Conference on Mechatronics and Computer Technology Engineering (MCTE 2022), 2022, Chongqing, China

Abstract

In order to improve the performance of automatic driving target detection in foggy scenarios, this paper proposes an end-to-end domain adaptive target detection method based on YOLOV5 algorithm. The domain adaptive network is added to the last layer of the feature extractor of YOLOV5 algorithm, and the gradient inversion layer (GRL) is added between the label classifier and the domain classifier by adversarial training strategy, which confuses the ability of the network to distinguish the source domain from the target domain and maximizes the feature extractor to generate domain-invariant features. According to the training results on the dataset Cityscapes, compared with directly using the original YOLOV5 model, the average accuracy of the proposed algorithm is improved by 10.3%, and the FPS of the model reaches 42.8. Compared with the two-stage detection model Faster-RCNN, it is more suitable for the real-time scene of automatic driving.

Citation Download Citation

Fan Zhang "Domain adaptive foggy object detection method based on YOLOv5", Proc. SPIE 12500, Fifth International Conference on Mechatronics and Computer Technology Engineering (MCTE 2022), 125006H (16 December 2022); https://doi.org/10.1117/12.2661003

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