Deep Learning Based Algorithm For Automatic Target Recognition

Automatic Target Recognition(ATR)technology is urgently needed in the military field,the photoelectric based detection method is one of the key detection approach of modern early warning detection network,in actual combat,by the use of photoelectric device on different weapons and equipment platform,one can obtain different types,different time and different resolution of massive images,videos,and it is significance to realize rapid and accurate military target recognition in the collected images,and videos.Thanks to big data,cloud computing,and the development of artificial intelligence technology,especially the progress of intelligent target recognition technology which is based on the deep learning algorithm,now,it is possible to use machine to analysis military target.Based on the above analysis,this paper carry out the research of automatic target recognition which based on deep learning,main research works are as follows:(1)This paper proposes a target recognition algorithm DRFCN which is based on deep learning.Different from traditional target identification model of pip styles,by the way of dense connection of convolution block,each layer's characteristics are shared by every layer,expression of characteristics has been greatly increased.The experimental results show that DRFCN has advantages in mean accuracy precision and model size,at the same time,DRFCN has great effect on solving gradient vanishing,gradient exploding.(2)After model construction of DRFCN,this paper simulated the DRFCN algorithm on the standard data set of PASCAL VOC.Secondly,this paper created an equivalent scale military dataset according to PASCAL VOC format,and the resources are from actual combat environment.In the end,we retrain the DRFCN model on the dataset for military using.(3)After the completion of the above theoretical research and experimental simulation,this paper comprehensively use of Flask,Bottle,Sqlite,Caffe,CUDA,C++,Python,Nsight-eclipse and so on to design and develop a system for military target recognition,the system has the function of automatic target recognition of picture and videos.In the end,this paper test the system according to different conditions.