Spectral-Spatial Hyperspectral Image Classification Using Extended Multi Attribute Profiles and Guided Bilateral Filter

Abstract

The combination of spectral and spatial information for classification of hyper spectral image is an effective way in improving classification accuracy. In the paper, we proposed a new spectral-spatial method for textural feature extraction based on morphological attribute profiles and guided bilateral filter. Firstly, we obtained multi-level characters through the cascade of many attribute profiles to present the spatial and spectral information of remote sensing image. Then, bilateral filter preserved the edges of features with guide of the segmentation image generated by entropy rate super pixel algorithm. Finally, a pixel-wise classifier, e.g., Support vector machine and sparse representation, is used for classification based on the features. Experiments of two benchmark hyper spectral data sets showed better performance of the proposed method than other state-of-the-art methods.