A bullet recognition algorithm based on improved YOLOX

Frontiers in Physics

Physics

Yu Fu¹  Hua Chen²  Jie Hu³  Yubo Xu⁴ 

[1] Norinco Group Testing and Research Institute, Huayin, China;School of Mechatronic Engineering, Xi’an, China;School of Mechatronic Engineering, Xi’an, China;Norinco Group Testing and Research Institute, Huayin, China;School of Optoelectronic Engineering, Xi’an, China;

关键词: target detection;    deep learning;    YOLOX;    attention mechanism;    loss function;   

DOI  :  10.3389/fphy.2023.1213489

 received in 2023-04-28, accepted in 2023-06-23,  发布年份 2023

来源: Frontiers

【 摘 要 】

The identification and positioning of flying projectiles is a critical issue that affects the testing accuracy of equipment in ballistic testing technology. Traditional image processing methods are difficult to accurately extract targets due to the characteristics of small target size, fast speed, and strong fragmentation interference of projectiles ejected from the muzzle. This paper proposes a projectile recognition algorithm based on an improved YOLOX detection model for the detection and recognition of flying projectiles. The fast and accurate YOLOX model is used, and the network structure is improved by adding attention mechanisms in the feature fusion module to enhance the detection ability of small targets; the loss function is also improved to enhance the model’s iterative optimization ability. Test results show that the improved YOLOX model has significantly improved accuracy in projectile recognition compared to the original network, reaching 84.82%, demonstrating the feasibility of the proposed approach. The improved algorithm can be effectively used for small target scenarios in range testing and significantly improves the accuracy of recognition.

【 授权许可】

Unknown   
Copyright © 2023 Hu, Chen, Xu and Fu.