Research on multi-target detection method based on deep learning

Kang Dai; Xiubao Sui; Liping Wang; Qiuhao Wu; Qian Chen; Guohua Gu

doi:10.1117/12.2587486

12 March 2021 Research on multi-target detection method based on deep learning

Kang Dai, Xiubao Sui, Liping Wang, Qiuhao Wu, Qian Chen, Guohua Gu

Proceedings Volume 11763, Seventh Symposium on Novel Photoelectronic Detection Technology and Applications; 117637U (2021) https://doi.org/10.1117/12.2587486
Event: Seventh Symposium on Novel Photoelectronic Detection Technology and Application 2020, 2020, Kunming, China

Abstract

Aiming at the problems of large recognition error and low detection accuracy of small targets in target detection, this paper proposes a multi-target detection algorithm based on deep learning. This method is improved on the basis of YOLOv3. First, the Darknet network is modified, and a set of residual blocks is added at the end of the network to obtain feature maps of 4 scales. Second, replace IOU with DIOU to increase network robustness. Finally, the K-means++ algorithm is used to generate 12 anchor boxes on the training set. The results show that when the input is 576x576, based on the Nvidia 980Ti graphics card, the average accuracy of this algorithm on the VOC 2007 data set reaches 80.1% and the speed reaches 52.64fps, which improves the detection accuracy of multi-target objects while ensuring real-time performance.

Citation Download Citation

Kang Dai, Xiubao Sui, Liping Wang, Qiuhao Wu, Qian Chen, and Guohua Gu "Research on multi-target detection method based on deep learning", Proc. SPIE 11763, Seventh Symposium on Novel Photoelectronic Detection Technology and Applications, 117637U (12 March 2021); https://doi.org/10.1117/12.2587486

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