基于深度学习的军事目标识别算法综述  

Overview of Military Target Recognition Algorithms Based on Deep Learning

作　　者：宋晓茹[1] 刘康 高嵩[1] 陈超波[1] SONG Xiao-ru;LIU Kang;GAO Song;CHEN Chao-bo(School of Electronic Information Engineering,Xi an Technological University,Xi an 710021,China)

机构地区：[1]西安工业大学电子信息工程学院,西安710021

出　　处：《科学技术与工程》2022年第22期9466-9475,共10页Science Technology and Engineering

基　　金：陕西省重点研发计划(2021GY-287);机电动态控制重点实验室开放课题基金(6142601200301)。

摘　　要：目标识别作为深度学习中最受欢迎的领域之一,已广泛应用于民用的各个方面,如人脸识别、行人重识别、车牌识别、车辆识别等;而在军事应用领域,由于军事目标数据集较少,但识别要求精度高实时性强,所以还在发展阶段。首先阐述了基于深度学习的军事目标识别发展现状;然后介绍了6种目前主流的基于深度学习的军事目标识别算法(包括Mask R-CNN、GAN与深度森林、DRFCN、E-MobileNet、SSD300、YOLO)及相关网络结构、改进方法与实际应用;最后对主流方法进行总结,并探讨了未来的发展趋势。As one of the most popular fields in deep learning, target recognition has been widely applied for civil use, such as face recognition, pedestrian recognition, license plate recognition, vehicle recognition and so on. In the field of military application, it is in its development stage given that there are few military target data sets, but the recognition requires high precision and strong real-time performance. Firstly, the development status of military target recognition was expounded based on deep learning. Then, six mainstream military target recognition algorithms were introduced on the basis of deep learning(including Mask R-CNN, GAN and Deep Forest, DRFCN, E-MobileNet, SSD300 and YOLO) and their related network structures, improvement methods and practical applications. Finally, the mainstream methods are summarized, and the future development trend was discussed.

关 键 词：深度学习 军事目标 高精度 实时性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]