摘要
文中提出一种基于CGAN和GcForest的军事目标识别方法,通过构建CGAN对军事目标样本进行扩展和质量提升。进而基于启发式学习进行抽样迭代,构建有效的样本训练集。在高质量训练集的基础上,通过GcForest进行有监督的学习,最终得到军事目标识别的分类网络模型。文中所提方法相对于CNN、KNN、SVM等方法,在目标样本识别整体准确率上高出7.80%~75.27%,同时在不同小规模样本的条件下整体准确度高出29.21%~67.50%。
A military target recognition method is proposed based on CGAN and GcForest.CGAN is constructed to expand and improve the quality of military target samples.Based on heuristic learning,sampling iteration is carried out to construct an effective sample training set.On the basis of high-quality training set,supervised learning is conducted through GcForest,and finally the classified network model of military target identification is obtained.Compared with CNN,KNN,SVM and other methods,the proposed method is 7.80%-75.27%higher in the overall accuracy of target sample identification,and 29.21%-67.50%higher in the overall accuracy of different small samples.
作者
林洋
董宝良
刘泽平
LIN Yang;DONG Bao-liang;LIU Ze-ping(No.15 Research Institute of China Electronics Technology Group Corporation,Beijing 100083,China)
出处
《信息技术》
2020年第3期134-138,共5页
Information Technology