Classification of Multispectral Satellite Images using Sparse SVM Classifier

This work proposes an efficient classification scheme for identifying various land classes present in a multispectral satellite image. This spectral image provides extensive knowledge about land cover mapping in multispectral satellite images. This paper proposes an efficient technique in land cover classification which involves fuzzy hybrid with hierarchical clustering applied then to the sparse SVM classifier. Initially preprocessing is done using Gaussian filter and transformed to a suitable form using Wavelet transform. Subsequently, segmentation is performed in the wavelet transformed image using fuzzy hybrid with hierarchical clustering technique. Then the proposed sparse SVM classifier is trained by the features obtained from the clustered output. Thus the multispectral image of various satellite images can be classified into different land classes comparing with the training data given to sparse SVM. The performance is evaluated by comparing with the existing classifiers for different multi-spectral satellite images which provides accurate results. The classification accuracy is measured from the performance analysis graph where the results demonstrate that the proposed sparse SVM classifier can optimally enhance the classification accuracy of any multispectral satellite image.