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.
作者
史国川
拓浩男
曹宇剑
王民
丁健
SHI Guochuan;TUO Haonan;CAO Yujian;WANG Min;DING Jian(Army Academy of Artillery and Air-Defence,Hefei 230031;Hefei Academy,Hefei 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。
关键词
无人机
目标检测
卷积神经网络
准确率
UAV(Unmanned Aerial Vehicle)
target detection
CNN(Convolution Neural Network)
accuracy