Monitoring land-cover changes by combining a detection step with a classification step

Abstract

An approach merging the HotellingT 2 control scheme with weighted random forest classifier is proposed and used in the context of detecting land cover changes via remote sensing and radiometric measurements. HotellingT 2 procedure is introduced to identify features corresponding to changed areas. However, T 2 scheme is not able to separate real from false changes. To tackle this limitation, the weighted random forest algorithm, which is an efficient classification technique for unbalanced problems, has been successfully applied on features of the detected pixels to recognize the type of change. The performance of the algorithm is evaluated using SZTAKI AirChange benchmark data, results show that the proposed detection scheme succeeds to appropriately identify changes to land cover. Also, we compared the proposed approach to that of the conventional algorithms (i.e., neural network, random forest, support vector machine and k-nearest neighbors) and found improved performance.