Lightweight small target detection algorithm based on YOLOv5

Nan Song; Ying Wang; Zhiyi Xin; Huili Xin; Tianxu Liu

doi:10.1117/12.2692005

23 August 2023 Lightweight small target detection algorithm based on YOLOv5

Nan Song, Ying Wang, Zhiyi Xin, Huili Xin, Tianxu Liu

Proceedings Volume 12784, Second International Conference on Applied Statistics, Computational Mathematics, and Software Engineering (ASCMSE 2023); 127843H (2023) https://doi.org/10.1117/12.2692005
Event: 2023 2nd International Conference on Applied Statistics, Computational Mathematics and Software Engineering (ASCMSE 2023), 2023, Kaifeng, China

Abstract

To address the problems of current small target detection algorithms such as slow detection speed, high memory consumption, and not well applied to mobile devices, an improved lightweight small target detection algorithm MTG-YOLOv5 is proposed based on YOLOv5s algorithm. firstly, MobileNet V3 is used instead of the backbone network to reduce the number of network parameters and increase the speed of operation. Secondly, the small target detection layer is added to the original network structure to enhance the detection capability of small targets. Finally, Ghost module is introduced to replace the commonly used convolutional kernel to ensure detection accuracy and further light weighting. The improved algorithm was tested on the processed TinyPerson dataset. The experimental results show that the overall average accuracy of mAP of YOLOv5s algorithm is improved by 4.5%, the parameters are reduced by 1/2, the FPS is improved from 123.54 to 187.62, and the speed is enhanced by 51.87%. The algorithm effectively improves the detection speed while ensuring the detection accuracy.

Citation Download Citation

Nan Song, Ying Wang, Zhiyi Xin, Huili Xin, and Tianxu Liu "Lightweight small target detection algorithm based on YOLOv5", Proc. SPIE 12784, Second International Conference on Applied Statistics, Computational Mathematics, and Software Engineering (ASCMSE 2023), 127843H (23 August 2023); https://doi.org/10.1117/12.2692005

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