Classification of high resolution hyperspectral remote sensing data using deep neural networks

The high resolution hyperspectral remote sensing data collected from urban and landscape areas have been extensively studied over the past decades. Recent applications pose an emerging need of analyzing the land cover types based on high resolution hyperspectral remote sensing data originating from remote sensory devices. Toward this goal, we propose a deep neural network (DNN) classifier in this paper. The DNN is constructed by combining a stacked autoencoder with desired numbers of autoencoders and a softmax classifier. Our experimental results based on the hyperspectral remote sensing data demonstrate that the presented DNN classifier can accurately distinguish different land covers including the mixed deciduous broadleaf natural forest and different land covers such as agriculture, roads, buildings, etc. We test the proposed method by using three different benchmark data sets. The proposed method showcases the huge potential of deep neural networks for hyperspectral data analysis.