Wavelet-Based Classification of Remotely Sensed Images: A Comparative Study of Different Feature Sets in an Urban Environment

Abstract

This paper presents a series of experiments on classification of remotely sensed images, to investigate the effectiveness of various combinations of different types of feature sets, including spectral features, variance features and wavelet-based features. All the experiments use the identical study area, training data, reference data, testing data, and classification algorithm while varying the feature sets. The classification accuracy from different feature sets is evaluated using the traditional accuracy assessment from reference data. The experimental results show that the spectral-based feature set has the basic discrimination power to distinguish classes with middle and high homogeneity value. However, it has little success in correctly classifying classes with low homogeneity value, such as the residential class. Compared with spectral features, the multi-scale wavelet-based feature set can improve the discrimination power for classes with both low and high homogeneity value. The variance-based feature set alone has little discrimination power, no matter what homogeneity level the class has. However, adding the variance features into the spectral feature or wavelet-based feature set can dramatically increase the classification accuracy for classes with low homogeneity value.

Keywords: Remote sensing, image classification, wavelet, land use, land cover, feature extraction