基于小样本学习的军事装备识别技术  

Military equipment recognition technology based on small sample learning

作　　者：靳文兵[1] 郭江宇[1] 陈志华 吕昊 黄明杰 JIN Wenbing;GUO Jiangyu;CHEN Zhihua;LYU Hao;HUANG Mingjie(North Automatic Control Technology Institute,Taiyuan 030006,China)

机构地区：[1]北方自动控制技术研究所,太原030006

出　　处：《兵器装备工程学报》2023年第7期32-38,共7页Journal of Ordnance Equipment Engineering

摘　　要：针对军事装备样本稀缺昂贵、目标识别研究进展缓慢等问题,从人类认知科学出发,提出利用民用装备数据完成深度学习训练,实现大样本知识积累;采用模式识别算法提取装备武器特征,实现样本目标识别。通过分析2种样本属性空间距离,利用卷积神经网络作为前端,采用SIFT算法作为后端,构建递进学习模型,实现对多种军事装备的高效识别。实验测试模型识别坦克、战机和军舰平均置信度分别为87%、92%和91%,军事装备武器样本数量决定目标识别精度,样本泛化可提高目标识别率。提出的递进学习模型充分利用深度学习和模式识别算法优势,实现军事领域武器装备小样本目标识别。In view of the scarcity and high cost of military equipment samples and the slow progress of target recognition research,starting from human cognitive science,this paper proposes to use civil equipment data to complete deep learning training and realize the knowledge accumulation of large samples.The pattern recognition algorithm is used to extract the characteristics of equipment and weapons,and the sample target recognition is realized.By analyzing the spatial distance of the attributes of the two samples,the convolutional neural network is used as the front end,and the SIFT algorithm is used as the back end to build a progressive learning model to realize efficient identification of various military equipment.The average confidence of tanks,aircraft fighters and warships identified by the experimental test model is 87%,92% and 91% respectively.The target identification accuracy is determined by the number of weapon samples of military equipment.The generalization of samples can also improve the target identification rate.The proposed progressive learning model makes full use of the advantages of deep learning and pattern recognition algorithms to realize small sample target recognition in the military field.

关 键 词：深度学习 小样本学习 目标识别 军事装备 武器特征 

分 类 号：TP302.1[自动化与计算机技术—计算机系统结构]