Overview of Hierarchical Models for Hyperspectral Image Classification

Hyperspectral imaging enables accurate classification, but also presents challenges of high-dimensional data analysis. While pixelwise classification methods classify each pixel independently, recent studies have shown the advantage of considering the correlations between spatially adjacent pixels for accurate image analysis. This paper provides an overview of the available hierarchical models for spectral-spatial classification of hyperspectral images. The two most recent models are experimentally compared on a 102-band ROSIS image of the Center of Pavia, Italy. The experimental results demonstrate that classification methods using hierarchical models are attractive for remote sensing image analysis.