A Unified Approach to Change Detection Using an Adaptive Ensemble of Extreme Learning Machines

Abstract

Change detection for remote sensing (RS) images is a challenging task. The variation in the spatial, radiometric, spectral, and temporal resolution of the images adds to the complexity of the change-detection process. The application domain also has an impact on the decision to use a particular change-detection technique. There is no generic classification algorithm, which can be used for different application domains using the RS images like green cover, land use, forest fires, and so on. This letter proposes an adaptive ensemble of extreme learning machines (ELMs) for classification of RS images into change/no-change classes. ELM has good generalization capability and a fast learning phase. Therefore, an adaptive ensemble of different ELMs has been proposed. The proposed algorithm has been implemented in five sets of data. It has been used for earth monitoring applications, viz. green cover, flat fires, water bodies, and so on. Here the results for two data sets, viz. Sardinia Island and Mexico Fire have been presented. The results thus obtained are highly promising. They show an average accuracy of 90.5% in detecting the change.