基于改进R-FCN框架的遥感影像飞机目标识别方法  

RS image aircraft target identification method based on improved R-FCN detection framework

作　　者：王冰 周焰[1] 张怀念 王宁 WANG Bing;ZHOU Yan;ZHANG Huainian;WANG Ning(Air Force EarlyWarning Academy,Wuhan 430019,China)

机构地区：[1]空军预警学院

出　　处：《空军预警学院学报》2019年第5期318-322,共5页Journal of Air Force Early Warning Academy

基　　金：国家自然科学基金资助项目(61601510)

摘　　要：遥感影像飞机目标识别是实现地面特定目标的精准打击、掌握机场军事价值的重要途径.针对飞机识别数据集未充分参照不同条件下飞机几何形态的问题,构建了飞机类型识别数据集,同时为进一步提高识别精度,基于区域全卷积网络(R-FCN)识别框架,提出飞机目标全卷积神经网络(AFFCN)识别方法.通过人工增强方法,扩增包含四种类型飞机影像的数量,构建了每种类型飞机识别数据集;基于深度残差网络能有效区分不同类型目标的性质,提出了飞机目标深度残差网络,并将此网络应用于R-FCN识别框架中,建立了AFFCN识别方法.仿真结果表明,该方法结合本文数据集可以准确地识别遥感影像中的飞机目标.Remote sensing(RS)image aircraft target identification is an important way to achieve a precise attack on specific targets on the ground and to grasp the military value of the airport.In order to solve the problem that aircraft type identification data sets do not fully refer to aircraft geometry under different conditions,and to further improve the identification accuracy,this paper constructs the aircraft type identification data set.Based on the regional full convolution network(R-FCN)identification framework,the paper proposes the aircraft full convolution neural network(AFFCN)identification method,through the artificial enhancement method,the number of images containing four types of aircraft is amplified,with identification data set for each aircraft type constructed.On the basis of the fact that deep residual network can effectively distinguish the properties of different types of targets,the paper puts forward the aircraft target depth residual network,which is applied to the R-FCN identification framework to establish the FCN recognition method.Simulation results show that the method combined with the data set presented in this paper can accurately identify aircraft targets in RS images.

关 键 词：遥感影像处理 飞机目标识别 飞机目标全卷积神经网络识别 飞机类型识别数据集 R-FCN框架 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置] TN957[自动化与计算机技术—控制科学与工程]