摘要
为了更好的对复杂战场环境下军事目标检测和侦查,提出一种基于YOLOv3-SPP的改进算法。通过收集不同目标尺寸、类别等条件下坦克、步战车、雷达等军事对象,构建军事目标小型数据集;对数据集进行数据增强处理,扩充样本数,提高训练模型鲁棒性;将DIoU和Focal Loss替换均方误差函数和交叉熵函数,提高目标检测算法精度;利用K-means++聚类算法计算得出适用的锚框,进一步提高模型检测精度。实验结果表明,改进的YOLOv3-SPP军事目标检测算法相对于原YOLOv3-SPP算法,模型收敛更快,平均精度提高了10%,精度和召回率分别提高了9%和8%,具备良好的检测能力,能为战场环境下军事目标的检测和侦查任务提供技术支持。
To better detect and investigate military targets in complex battlefield environments,this paper proposes an improved algorithm based on YOLOv3-SPP.By collecting military objects such as tanks,infantry fighting vehicles and radars of different target sizes and categories,a small data set of military targets is constructed.Then,data enhancement is performed on the data set,the number of samples is expanded,and the robustness of the training model is improved.D IoU and Focal Loss are used to replace the mean square error function and the cross entropy function to improve the accuracy of the target detection algorithm.The K-means++clustering algorithm is used to calculate the applicable anchor frame,which further improves the model detection accuracy.The experimental results show that,compared with the original YOLOv3-SPP algorithm,the improved YOLOv3-SPP military target detection algorithm has faster model convergence,with 10%higher average precision,9%higher precision and 8%higher recall rate.With good detection ability,it can provide technical support for the detection and reconnaissance tasks of military targets in the battlefield environment.
作者
洪毕辉
李文彬
朱炜
王晓鸣
张克斌
HONG Bihui;LI Wenbin;ZHU Wei;WANG Xiaoming;ZHANG Kebin(National Defense Key Subject Laboratory of Intelligent Ammunition Technology,Nanjing University of Science and Technology,Nanjing 210003,China)
出处
《兵器装备工程学报》
CAS
CSCD
北大核心
2023年第4期268-274,共7页
Journal of Ordnance Equipment Engineering
基金
青年科学基金项目(12102198)。
