Improved hyperspectral image classification with noise reduction pre-process

This paper shows that hyperspectral image classification performance using support vector machines (SVM) and relevance vector machines (RVM) can significantly be improved using a noise reduction pre-process. A wavelet domain, spatially adaptive denoising method that estimates the probabiliy that a coefficient represents a significant noise-free component is used for denoising of hyperspectral images before classification. It is shown that support vector machine and relevance vector machine classification of denoised hyperspectral images gives significantly better classification accuracy and furthermore improves sparsity.