基于中心损失函数的小样本SAR图像识别方法  

Small Sample SAR Image Recognition Method Based on Central Loss Function

作　　者：毛轩昂 刘振国[1] 姚陈芳 MAO Xuanang;LIU Zhenguo;YAO Chenfang(North Automatic Control Technology Institute,Taiyuan 030006,China;Central Reserve Information and Communication Brigade,Taiyuan 030000,China)

机构地区：[1]北方自动控制技术研究所,太原030006 [2]战略支援部队中部预备役信息通信大队,太原030000

出　　处：《火力与指挥控制》2024年第5期172-178,183,共8页Fire Control & Command Control

摘　　要：提出了一种基于中心损失函数的监督学习方法,用于改善小样本下的合成孔径雷达(synthetic aperture radar,SAR)图像识别性能。该方法通过学习每个类别的类别中心,并惩罚样本的深度特征与其相应类别中心之间的距离,从而提高类间分离度和类内分散度。为了验证方法的有效性,将所提方法与常见的深度学习算法在MSTAR图像识别数据集上进行比较。实验结果表示,相较于其他深度学习模型,该方法在小样本情况下有着更为卓越的图像识别性能。A supervision learning method based on the central loss function has been proposed to enhance the recognition performance of Synthetic Aperture Radar(SAR)images in small-sample scenarios.This method involves learning category centers for each class and penalizing the distance between the deep features of samples and their respective category centers,thereby improving both interclass resolution and intra-class dispersity.To validate the effectiveness of this approach,it is compared with common deep learning algorithms on the MSTAR image recognition dataset.The experimental results show that,compared to other deep learning models,this method exhibits more superior image recognition performance in scenarios with small samples.

关 键 词：合成孔径雷达 小样本图像识别 中心损失函数 深度学习 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]