TY - GEN
T1 - An Improved Lightweight YOLOv5 Network for Small Targets Detection
AU - Cai, Zhihao
AU - Luo, Xiangjie
AU - Zhao, Jiang
AU - Wang, Yingxun
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.
PY - 2025
Y1 - 2025
N2 - With the continuous development of unmanned systems, Unmanned aerial Vehicles (UAVs) have been paid more and more atten tion because of their unique advantages in the air. Target detection replaces the traditional human eye detection, which has the advantage of being faster and more accurate. However, the current UAV target detec tion also has some problems: for example, the target is too small and the detection algorithm is too complex. Based on the existing YOLOv5 net work framework, we propose a lightweight model for small target detec tion. In this paper, a fourth feature extraction channel is added on the basis of the original YOLOv5 three-scale feature detection network. The feature map size in this channel is 160× 160, which is especially for small target detection. In order to lighten the network and reduce the com putation, we introduced Ghost convolution module in YOLOv5 neck network. The improved model is tested on the VisDrone2019 dataset, and the results show that compared with the original YOLOv5 algo rithm, the proposed algorithm achieves higher detection accuracy with less computation and model complexity.
AB - With the continuous development of unmanned systems, Unmanned aerial Vehicles (UAVs) have been paid more and more atten tion because of their unique advantages in the air. Target detection replaces the traditional human eye detection, which has the advantage of being faster and more accurate. However, the current UAV target detec tion also has some problems: for example, the target is too small and the detection algorithm is too complex. Based on the existing YOLOv5 net work framework, we propose a lightweight model for small target detec tion. In this paper, a fourth feature extraction channel is added on the basis of the original YOLOv5 three-scale feature detection network. The feature map size in this channel is 160× 160, which is especially for small target detection. In order to lighten the network and reduce the com putation, we introduced Ghost convolution module in YOLOv5 neck network. The improved model is tested on the VisDrone2019 dataset, and the results show that compared with the original YOLOv5 algo rithm, the proposed algorithm achieves higher detection accuracy with less computation and model complexity.
KW - Lightweight
KW - Small target detection
KW - UAVs
KW - YOLOv5
UR - https://www.scopus.com/pages/publications/105006419989
U2 - 10.1007/978-981-96-2208-5_3
DO - 10.1007/978-981-96-2208-5_3
M3 - 会议稿件
AN - SCOPUS:105006419989
SN - 9789819622078
T3 - Lecture Notes in Electrical Engineering
SP - 22
EP - 31
BT - Advances in Guidance, Navigation and Control - Proceedings of 2024 International Conference on Guidance, Navigation and Control Volume 3
A2 - Yan, Liang
A2 - Duan, Haibin
A2 - Deng, Yimin
PB - Springer Science and Business Media Deutschland GmbH
T2 - International Conference on Guidance, Navigation and Control, ICGNC 2024
Y2 - 9 August 2024 through 11 August 2024
ER -