摘要
针对遥感影像中目标方向、目标大小、拍摄角度及场景的多样性导致飞机目标检测精度不高的问题,提出一种基于残差网络(ResNet)的目标检测新方法。首先采集并且标注遥感图像数据,这些数据包含了晴天、薄雾等多种气候条件下的遥感影像;然后构建图形金字塔和模板金字塔进行多尺度检测,并且加入残差网络的全卷积网络结构中不同层的上下文特征信息;最后通过拟合回归进行端到端的训练,得出鲁棒性强,精度高的目标检测网络模型。实验结果表明,该网络模型对于较复杂背景等干扰有较强的鲁棒性,检测精度高达89.5%。
The target in remote sensing image with different size and directions,also with different shot angles and complex background is difficult to detect.An algorithm based on residual network is proposed to solve these problems.Firstly,collection and annotation of remote sensing image data sets,which include the remote sensing images under a variety of climatic conditions such as sunny days and mist;then,the image pyramid and the template pyramid are constructed for multi-scale detection,and the contextual information of different layers in the full convolutional network structure of the residual network is added;finally,a good performance model trained by a end-to-end training weights with regression method.Experimental results show that the proposed method has high robustness against more complicated background disturbances,and precision is 89.5%.
作者
赵丹新
孙胜利
ZHAO Dan-xin;SUN Sheng-li(Chinese Academy of Sciences,Shanghai Institute of Technical Physics,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China;School of Information Science&Technology,ShanghaiTech University,Shanghai 201210,China;Key Laboratory of Infrared System Detection and Imaging Technology of the Chinese Academy of Sciences,Shanghai 200083,China)
出处
《电子设计工程》
2018年第22期164-168,共5页
Electronic Design Engineering