一种结合非顶层特征图和自适应阈值的飞机目标检测算法  

A Combination of Non-Top-Level Feature Maps and Adaptive Thresholds Aircraft Target Detection Algorithm

作　　者：谭振宇 江刚武[1] 刘建辉[1] TAN Zhenyu;JIANG Gangwu;LIU Jianhui(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学

出　　处：《测绘科学技术学报》2019年第4期382-387,共6页Journal of Geomatics Science and Technology

摘　　要：针对Faster R-CNN模型在遥感影像中对飞机目标进行检测与识别时,易出现漏检、错检等问题,提出了一种在基于小样本飞机遥感影像数据集的改进型Faster R-CNN目标检测方法。首先对特征提取网络进行优化,然后将非顶层特征图融合顶层特征图得到边缘信息更丰富的融合特征图,利用该特征图输入RPN网络,完成目标检测模型的训练;并结合自适应阈值进行检测。以普通客机与战斗机目标为试验对象,开展飞机目标检测与识别对比分析。试验结果表明,所提出的算法在小样本情况下检测效果有明显提升。When the aircraft target is detected and identified in the remote sensing image for the Faster R-CNN model,it is prone to problems such as missed detection and misdetection.An improved Faster R-CNN target detection method based on small sample aircraft remote sensing image dataset is proposed.Firstly,the feature extraction network is optimized,and then the non-top feature map is merged with the top feature map to obtain the rich feature of edge information.The figure is input into the RPN network by using the feature map to complete the training of the target detection model,and is combined with the adaptive threshold for detection.The targets of ordinary passenger aircraft and fighters are taken as the test object,and the comparative analysis of aircraft target detection and recognition is carried out.The test results show that the proposed algorithm has a significant improvement in the detection of small samples.

关 键 词：飞机 目标检测 FASTER R-CNN模型 RPN网络 特征图 

分 类 号：P237[天文地球—摄影测量与遥感]