摘要
针对遥感影像场景复杂,飞机目标尺寸小、特征不明显的问题,提出一种基于改进YOLOv3的遥感影像飞机目标检测算法。首先对YOLOv3的特征提取网络的结构进行改进,并将网络的检测尺度由3个扩展至4个,提高小目标的检测率;其次采用线性加权的非极大值抑制算法,降低排列交错紧密的小目标的漏检率;最后在本文设计的数据集上将该算法与YOLOv3进行对比实验。结果表明,改进后的算法对复杂背景下的小尺寸飞机目标的检测准确率和召回率均有明显提升,验证了本文算法的有效性和鲁棒性。
Aiming at the problem that the scene of remote sensing image is complicated and the size of the aircraft target is small,the characteristics of the aircraft target are not obvious,an aircraft target detection algorithm based on improved YOLOv3 is proposed in this paper.Firstly,the network structure of YOLOv3’s feature extraction network is improved and the detection scale of the network is extended from 3 to 4 to improve the detection rate of small targets.Secondly,a linearly weighted non-maximum suppression algorithm is used to reduce the missed detection rate of closely spaced small targets.Finally,this algorithm is compared with YOLOv3 on the data set designed in this paper.The results show that the improved algorithm can significantly improve the detection accuracy and recall rate of small-sized aircraft targets under complex backgrounds,which validates the effectiveness and robustness of the algorithm.
作者
袁铭阳
姜挺
王鑫
YUAN Mingyang;JIANG Ting;WANG Xin(Information Engineering University,Zhengzhou 450001,China)
出处
《测绘科学技术学报》
北大核心
2019年第6期614-619,共6页
Journal of Geomatics Science and Technology