Land Cover Mapping using Adaptive Decision Tree Algorithm for WorldView-2 High-Resolution Images

Recently, there is a massive development in remote sensing applications due to the availability of various resources of high spatial and spectral resolutions. The motivation of this research is to get the maximum benefits of high spectral resolution satellite images using a robust and fast classification algorithm. Firstly, Top of Atmosphere (TOA) correction is applied on WorldView-2 (WV-2) images to remove the effects of variations relative to sensor error and positions of the sun, earth, and satellite. Then, the TOA corrected image is classified using an adaptive decision tree algorithm. The proposed algorithm uses indices with adaptive thresholds to generate a land cover map. Adaptive thresholds are calculated automatically according to input samples from the study area. The proposed algorithm successfully maps different areas rather than using fixed thresholds for a specific area. The classification results of the proposed algorithm are compared against those of the maximum likelihood classification (MLC) technique. Assessment for the proposed algorithm is achieved using overall accuracy and kappa coefficient calculated from the confusion matrix. The overall accuracy and kappa coefficient of the proposed algorithm are 92% and 0.90, respectively, for Brisbane city and 95% and 0.89, respectively, for Rio De Janeiro city. The proposed algorithm is simple and semi-automated and it is suitable for any image with the same spectral bands as the WV-2 satellite image.