Sparse smoothing preprocessing of hyperspectral images for improved classification performance

摘要

In this article, we propose to use the L-0 gradient minimization (L-0 GM) sparse smoothing technique as an image preprocessing method for hyperspectral data classification. This method can enhance the fundamental image constituents while diminishing insignificant details in the images. We performed experiments and provided a comparative analysis on a real benchmark hyperspectral scene with two classical classifiers (k-nearest neighbour classifier and support vector machine classifier). The experimental results show that the L-0 GM smoothing is a potential preprocessing technique that can effectively improve the classification performance.