Convolutional Neural Network-based Automatic Target Recognition Algorithm in SAR Image

被引：1

作者：

Cho, Jun Hoo ^{[1
]}

Kang, Chang Ho ^{[1
]}

Park, Chan Gook ^{[1
]}

机构：

[1] Department of Mechanical and Aerospace Engineering, Seoul National University, Korea, Republic of

来源：

Journal of Institute of Control, Robotics and Systems | 2017年 / 23卷 / 08期

关键词：

Classification (of information) - Automatic target recognition - Radar target recognition - Convolution - Neural networks - Image recognition - Radar imaging;

D O I：

10.5302/J.ICROS.2017.17.0100

中图分类号：

学科分类号：

摘要：

In this paper, we have designed an SAR automatic target recognition (SAR ATR) algorithm using the convolutional neural network (CNN), which has excellent image recognition performance. Previous SAR ATR methods are difficult to implement, because they include additional preprocessing processes or need prior SAR image information. To address these issues, we propose a CNN structure that is specialized for SAR image classification by modifying the structure of VGGNet. It is confirmed by simulation that the classification accuracy of the proposed method on the MSTAR SAR dataset is increased by 1-2% compared with the conventional VGGNet. Moreover, the classification performance is further improved when the train data is much smaller than the test data. © ICROS 2017.

引用

页码：644 / 649