A Spectral–Spatial Classification Algorithm for Multispectral Remote Sensing Data

Abstract

This paper aims at achieving improved land cover classification performance over conventional per-pixel classifiers as well as spectral-spatial classifiers such as ECHO (Extraction and Classification of Homogeneous Objects) algorithm. The proposed algorithm is a two-stage process, which makes use of the contextual information from neighboring pixels. First, a spatial filter is used to achieve more homogeneous regions. Secondly, maximum likelihood pixel classifier is employed to classify the land covers. The experimental results indicate that improved classification accuracy and smoother (more acceptable) thematic maps are achieved than what is obtained with the other methods considered.