Hyperspectral Remote Sensing Image Chassification Using the Stacked Sparse Autoencoder

To extract rich features of hyperspectral image, this study explores the deep features of the raw data by using a stacked sparse autoencoder in the deep learning theory. First we create a sparse expression of raw hyperspectral image using sparse autoencoder. Then a deep neural network generating the deep features of raw data is built through learning stacked sparse autoencoder layer by layer. In addition, the deep feature-related model parameters are precisely calibrated by the statistical learning algorithm of the support vector machine (SVM). The performance of the experiment indicates that the overall accuracy of classification model based on stacked sparse autoencoder reaches 87.82%, superior to other experimental methods. From our experiments, it follows that the deep learning theory and stacked sparse autoencoder are of high potential in hyperspectral remote sensing image classification.