Regression tree algorithm for classification of fused multispectral and panchromatic image

In this paper, classification of satellite image is done to detect vegetation, water, soil, built-up area etc. present in the region of which satellite image is captured. For classification regression tree algorithm is used. Regression tree algorithm uses threshold to detect class of data that gives least mean square error. This threshold is applied to NDVI, NDWI, SAVI, and NDBI to detect vegetation, water, soil and built-up area respectively. For classification Landsat 7 ETM+ multispectral images are used. But they have low spatial resolution. Hence they are fused with panchromatic image which is of high resolution. Image fusion is done using HIS transform, Brovey transform, PCA method, HPF method and wavelet transform. Fused image is given as input to regression tree for classification. Using this method various areas can be detected effectively as compared to the original satellite image. Classified image is compared with reference data to check the accuracy of classified image. It is observed that image fused using HPF method when classified using regression tree algorithm gives more effective results.