Deep convolutional recurrent neural network with transfer learning for hyperspectral image classification

• DOI: 10.1117/1.jrs.12.026028
• Publication Date: 2018-06-11T15:36:42Z
• Authors: Bing Liu; Xuchu Yu; Anzhu Yu; Gang Wan
• Keywords: Transfer of learning; Contextual image classification

The deep learning methods have recently been successfully explored for hyperspectral image classification. However, it may not perform well when training samples are scarce. A transfer method is proposed to improve the classification performance in situation of limited samples. First, a Siamese network composed two convolutional neural networks designed local descriptors extraction. Subsequently, pretrained model reused knowledge tasks by feeding features extracted from each band into recurrent network. Indeed, constructed this way. Finally, entire tuned small number labeled important characteristic that provides way utilizing spatial–spectral without dimension reduction. Furthermore, an opportunity train such with Experiments on three widely used datasets demonstrate can and competitive results be achieved compared state-of-the-art methods.