摘要
YOLOv5模型对普通场景图像的目标检测有更好的性能,但在高空航拍图像检测中表现不佳,针对这个问题,提出一种改进的YOLOv5模型。首先,建立高空航拍目标数据集,弥补该类图像不足的问题,对模型进行针对性训练,其次,采用多尺度细节增强提升处理数据图像,整体提升数据质量;最后,利用多尺度特征融合更好的平衡目标特征和位置信息,增加大尺度检测头提升小目标检测能力。经过实验分析,证明该方法在对高空航拍图像目标进行检测时平均精度、准确率和召回率分别比YOLOv5模型提高了12.6%、10.3%和6%,满足检测要求。
YOLOv5 has better performance in object detection of common scene images,but it has a poor performance in detecting objects in high-altitude aerial images.Aiming at this problem,this paper proposes an improved YOLOv5 model.Firstly,a target dataset of high-altitude aerial images is built to make up for the shortage of such images to train the model.Secondly,multi-scale detail enhancement is used to process data images to improve the overall quality of the data.Finally,multi-scale feature fusion is used to better balance the object features and location information,and the large-scale detection head is added to improve the small object detection ability.The experimental results show that the average accuracy,accuracy rate and recall rate of this method are 12.6%,10.3%and 6%higher than that of the YOLOv5 model respectively,which meets the detection requirements.
作者
谢晓竹
卢罡
XIE Xiaozhu;LU Gang(Army Academy of Armored Forces,Beijing 100072,China)
出处
《兵器装备工程学报》
CAS
CSCD
北大核心
2023年第1期248-253,共6页
Journal of Ordnance Equipment Engineering
基金
军内科研项目(LJ20202AXXX)。
