A 3-Stage Spectral-Spatial Method for Hyperspectral Image Classification

Hyperspectral images often have hundreds of spectral bands different wavelengths captured by aircraft or satellites that record land coverage. Identifying detailed classes pixels becomes feasible due to the enhancement in and spatial resolution hyperspectral images. In this work, we propose a novel framework utilizes both information for classifying The method consists three stages. first stage, pre-processing Nested Sliding Window algorithm is used reconstruct original data enhancing consistency neighboring then Principal Component Analysis reduce dimension data. second Support Vector Machines are trained estimate pixel-wise probability map each class using from Finally, smoothed total variation model applied ensure connectivity classification smoothing tensor. We demonstrate superiority our against state-of-the-art algorithms on six benchmark datasets with 10 50 training labels class. results show gives overall best performance accuracy even very small set labeled pixels. Especially, gain respect other increases when number decreases, and, therefore, more advantageous be problems set. Hence, it great practical significance since expert annotations expensive difficult collect.