Fusion of feature selection and optimized immune networks for hyperspectral image classification of urban landscapes

Abstract

The urban landscape is dynamic and complex. As improved remote sensing data in terms of spatial and spectral characteristics became available, more sophisticated methods have been adopted for urban applications. This study proposed and evaluated a classification model incorporating feature selection, artificial immune networks and parameter optimization. Information gain, a broadly applied feature selection metric used in data mining techniques such as decision trees, was used for feature selection. Two types of information gain binary-class entropy and multiple-class entropy - were investigated. Artificial immune networks have been recently applied to remote sensing classification and have been proven useful especially when multiple parameters of the networks are optimized through a genetic algorithm. The proposed model was tested for urban classification using hyperspectral (i.e. AISA and Hyperion) and LiDAR data over two urban study sites. Results show that the model considerably reduced processing time (similar to 70%) for classification without significant accuracy decrease.