基于人眼视觉皮层系统的SAR图像目标识别

SAR Target Recognition Based on Human Visual Cortical System

摘要:

人眼的视觉系统是已知最复杂最精确的目标识别系统；针对现有浅层学习模型提取SAR图像目标特征效果不理想、普适性不强及目标识别精度不高等问题,参考人眼视觉系统的分层特征提取机制提出了一种深层特征提取目标识别算法；算法首先提出一种针对SAR图像的预处理方法,在特征提取阶段应用多层稀疏自动编码器,通过逐层减少神经元数目使编码器学习得到输入数据的压缩表示,从而提取得到更抽象更鲁棒的目标特征；采用MSTAR数据库3个目标进行实验,试验结果表明,算法能很好地对MSTAR目标进行特征提取和识别,类比原有的方法有较高的识别准确率,达到96.7%。具有较强的实用性。

Abstract:

A method for SAR image target recognition based on human visual cortical system is presented in this paper since human Vision system is the most complex and accurate recognition system. First a new SAR Image preprocessing method is proposed. Then a sparse autoencoder is used to extract useful features by limit the number of cells in each layer. Experiment result of MSTAR public data show a better performance of recognition.