摘要
针对复杂光照环境下无人机作战的小目标难以检测问题,以引入注意力机制的YOLOv5s-Se模型为基础,设计了基于Ghost模块与形状损失函数的YOLOv5s-Se_Point改进模型。该模型首先引入形状特征损失函数,提高对小目标形状特征的提取能力。然后,使用Ghost模块替换网络Backbone中的卷积模块,以提高识别速度,降低模型尺寸。通过设置模拟作战场景进行实验,结果表明YOLOv5s-Se_Point相对于YOLOv5s和YOLOv5s-Se模型,在准确率和检测速度上具有一定优势,可以有效地改善复杂光照环境下传统算法的缺陷,提高算法的鲁棒性,实现对复杂光照环境下小目标的有效识别。
Aiming at small target detection in complex lighting environment during UAV combat,based on the YOLOv5s-Se model attention mechanism,an improved YOLOv5s-Se_Point model based on the Ghost module and shape loss function is proposed.Firstly,in order to improve the extraction ability of small targets’shape features,a shape loss function is introduced.Then the Backbone of the model is modified by replacing Conv Module with Ghost Module to improve the recognition speed and reduce the size of the model.Experiments are performed by setting up simulated combat scenarios,experimental results show that YOLOv5s-Se_Point has certain advantages in accuracy and detection speed compared with YOLOv5s and YOLOv5s-Se models,which can effectively improve the defects and the robustness of traditional algorithms,and achieve effective detection of small targets in complex lighting environments.
作者
郝立
张皓迪
HAO Li;ZHANG Haodi(School of Automation,Southeast University,Nanjing 210096,China;Chien-shiung Wu College,Southeast University,Nanjing 211189,China)
出处
《系统仿真技术》
2022年第2期85-89,95,共6页
System Simulation Technology
