基于Faster R-CNN的无人机侦察目标检测方法  

UAV Reconnaissance Target Detection Method Based on Faster R-CNN

作　　者：史国川 拓浩男 曹宇剑 王民 丁健[2] SHI Guochuan;TUO Haonan;CAO Yujian;WANG Min;DING Jian(Army Academy of Artillery and Air-Defence,Hefei 230031;Hefei Academy,Hefei 230601)

机构地区：[1]陆军炮兵防空兵学院,合肥230031 [2]合肥学院,合肥230601

出　　处：《舰船电子工程》2020年第4期37-39,87,共4页Ship Electronic Engineering

基　　金：安徽高校省级自然科学研究项目(编号:KJ2017A528)资助。

摘　　要：目标检测的准确率是评估侦察目标性能的重要指标之一。论文提出了基于Faster R-CNN的无人机侦察目标检测方法,对模型的RPN模块、多任务损失函数、卷积特征共享等算法进行了分析和研究,选取油库、舰艇、立交桥、飞机等四种典型目标,以Faster R-CNN为基准模型进行训练和测试,模型平均准确率为89.47%,目标检测准确率高。Target detection accuracy is one of the important indicators of reconnaissance goal of performance evaluation.In this paper,UAV reconnaissance target detection method is proposed based on faster R-CNN,algorithm of the model of RPN module,mul⁃titasking loss function,shared convolution characteristics are analyzed and researched,selection of four kinds of typical targets such as Oil depot,ships,overpass,aircraft,with faster R-CNN as a benchmark model for training and testing,the average accuracy of mod⁃el is 89.47%,target detection accuracy is high.

关 键 词：无人机 目标检测 卷积神经网络 准确率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]