| 摘要: |
| 近年来，卷积神经网络(CNN)已广泛应用于合成孔径雷达(SAR)目标识别。由于SAR目标的训练数据集通常较小，基于CNN的SAR图像目标识别容易产生过拟合问题。生成对抗网络(GAN)是一种无监督训练网络，通过生成器和鉴别器两者之间的博弈，使生成的图像难以被鉴别器鉴别出真假。本文提出一种基于改进的卷积神经网络(ICNN)和改进的生成对抗网络(IGAN)的SAR目标识别方法，即先用训练样本对IGAN进行无监督预训练，再用训练好的IGAN鉴别器参数初始化ICNN，然后用训练样本对ICNN微调，最后用训练好的ICNN对测试样本进行分类。MSTAR实验结果表明，提出的方法不仅能够在训练样本数降至原样本数30%的情况下获得高达96.37%的识别率，而且该方法比直接采用ICNN的方法具有更强的抗噪声能力。 |
| 关键词: 卷积神经网络 生成对抗网络 合成孔径雷达 目标识别 |
| DOI：DOI:10.3969/j.issn.1672-2337.2020.03.009 |
| 分类号:TN957.5 |
| 基金项目:国家自然科学基金(No.61501210)； 江西省自然科学基金(No.20161BAB202054)； 江西省教育厅科技项目(No.GJJ170825) |
|
| SAR Target Recognition Method Based on ICNN and IGAN |
|
CANG Mingjie， YU Lingjuan， XIE Xiaochun
|
|
1. School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, China;2. School of Physics and Electronic Information, Gannan Normal University, Ganzhou 341000, China
|
| Abstract: |
| In recent years, convolutional neural network (CNN) has been widely used in synthetic aperture radar (SAR) target recognition. Due to the small training dataset of SAR target, CNN-based SAR target recognition is prone to produce over-fitting. Generative adversarial network (GAN) is an unsupervised training network, where the game between generator and discriminator makes it difficult for the discriminator to identify the authenticity of the generated image. A method based on the improved convolutional neural network (ICNN) and the improved generative adversarial network (IGAN) is proposed in this paper for SAR target recognition. Firstly, IGAN is performed the unsupervised pre-training by training samples, and then ICNN is initialized according to the trained discriminator parameters of IGAN. After that, training samples are utilized to finely tune ICNN. Finally, the trained ICNN is used to classify the testing samples. MSTAR experimental results show that the proposed method can obtain a recognition rate up to 96.37% even when the number of training samples is reduced to 30%, and this method has stronger anti-noise capability than the method using ICNN directly. |
| Key words: convolutional neural network generative adversarial network synthetic aperture radar target recognition |
