Abstract. Terrain classification is an important application for polarimetric synthetic radar (PolSAR) image processing. Inspired by the popular deep belief network (DBN), a PolSAR classification method is proposed, which is called multilayer Wishart restricted Boltzmann machine (MWRBM). For PolSAR data, the traditional DBN is limited by binary value distribution, which is not suitable for PolSAR image classification. Therefore, according to the statistical distribution of PolSAR data, a new type of Wishart-restricted Boltzmann machine is proposed. An MWRBM, as one of the deep learning models, is proposed for PolSAR image classification. For improving the classification result, the labeled samples are used to fine-tune the parameters of the proposed deep model. Finally, two real PolSAR datasets are tested to verify the effectiveness of the proposed method. Experimental results demonstrate that the proposed method is very effective and compare favorably to the state-of-the-art methods.

中文翻译：

---

基于多层Wishart限制玻尔兹曼机的极化合成孔径雷达图像分类

摘要。地形分类是极化合成雷达 (PolSAR) 图像处理的重要应用。受流行的深度信念网络（DBN）的启发，提出了一种PolSAR分类方法，称为多层Wishart受限玻尔兹曼机（MWRBM）。对于 PolSAR 数据，传统 DBN 受限于二进制值分布，不适合 PolSAR 图像分类。因此，根据PolSAR数据的统计分布，提出了一种新型的Wishart限制玻尔兹曼机。MWRBM 作为深度学习模型之一，被提出用于 PolSAR 图像分类。为了改善分类结果，标记样本用于微调所提出的深度模型的参数。最后，测试了两个真实的 PolSAR 数据集以验证所提出方法的有效性。实验结果表明，所提出的方法非常有效，并且与最先进的方法相比具有优势。