Comparative study of classification methods for surficial materials in the Umiujalik Lake region using RADARSAT-2 polarimetric, Landsat-7 imagery and DEM Data

A study focusing on the classification of surficial materials in the Umiujalik Lake area using multisource data including polarimetric SAR data, Landsat optical data, and DEM has been conducted. The purpose of this study is to explore improving classification performance by comparing different feature combinations and different classifiers. First, four classification methods were compared on different combinations of features of intensity and texture. Second, the effects of dimension reduction algorithms for classification were investigated. Finally, six different dimension reduction methods were used to see if they can improve or remain classification performance by using fewer dimensions. Results show that adding texture features can help improve classification accuracy; the best classification accuracy is achieved by rotation forest classification method using the combination of intensity and texture features; the classification performance remains stable using fewer features.