Spectral-spatial classification of hyperspectral images based on joint bilateral filter and stacked sparse autoencoder

Abstract

Reducing noise in hyperspectral image (HSI) while preserving the details and extracting useful spectral-spatial information have always been one of the important problems of the classification task. This paper proposes a method by combining joint bilateral filter (JBF) and stacked sparse autoencoder via an ensemble strategy for the HSI classification. First, the novel JBF has an ability to preserve the important texture features and to restore the corrupted pixel, while extracting spectral and spatial information from hyperspectral data due to consider spectral as well as the spatial closeness for smoothing the HSI simultaneously. Second, stacked sparse autoencoderand (SSA) is adopted to adaptively extract high-level spectral-spatial feature representations from the smoothed image. Finally, the random forest (RF) classifier is adopted to perform supervised fine-tuning and classification. Experimental results on two real hyperspectral data sets demonstrate the effectiveness of the proposed classification approach.