Using Combination Technique for Land Cover Classification of Optical Multispectral Images

The need for efficient planning of the land is exponentially increasing because of the unplanned human activities, especially in the urban areas. A land cover map gives a detailed report on temporal dynamics of a given geographical area. The land cover map can be obtained by using machine learning classifiers on the raw satellite images. In this work, the authors propose a combination method for the land cover classification. This method combines the outputs of two classifiers, namely, random forests (RF) and support vector machines (SVM), using Dempster-Shafer combination theory (DSCT), also called the theory of evidence. This combination is possible because of the inherent uncertainties associated with the output of each classifier. The experimental results indicate an improved accuracy (89.6%, kappa = 0.86 as versus accuracy of RF [87.31%, kappa = 0.83] and SVM [82.144%, kappa = 0.76]). The results are validated using the normalized difference vegetation index (NDVI), and the overall accuracy (OA) has been used as a comparison basis.