空天预警研究学报2023，Vol.37Issue(4)：295-300,6.DOI:10.3969/j.issn.2097-180X.2023.04.012
一种深度神经网络SAR图像目标识别可视化方法
A visualization method for SAR image target recognition by deep neural network
马超 1王建明 1高华 2刘嘉铭1
作者信息
- 1. 南京电子技术研究所,南京 210039
- 2. 93534部队,江西景德镇 333012
- 折叠
摘要
Abstract
Aiming at the problem that the construction basis of discriminant model of deep neural network(DNN)model is not clear,this paper proposes a feature defect-based method of radar SAR image target recogni-tion and analysis.The paper uses the method to remove the local feature information of the target image,takes it as the input sample to obtain the variation relation of the recognition result of the DNN,and finally accords to the DNN model to classify the variation of the output results,analyzing the principle of DNN achieving target classifi-cation.Experiments conducted on publicly available MSTAR datasets show that the proposed method can effec-tively present the influence of different regions of the target information on deep network recognition,and realize the visualization analysis of DNN recognition.关键词
深度神经网络/SAR图像/雷达目标识别/特征缺损/可视化分析/滑动平均法Key words
deep neural network(DNN)/SAR image/radar target recognition/feature defect/visualization analysis/sliding-average algorithm分类
信息技术与安全科学引用本文复制引用
马超,王建明,高华,刘嘉铭..一种深度神经网络SAR图像目标识别可视化方法[J].空天预警研究学报,2023,37(4):295-300,6.
