Digital Assessment of Spatial Distribution of the Surface Soil Types Using Spatial (Texture) Features with MLC and SVM Approaches

In the present work, the effort has been made to identify and distribute of surface soil types using high spatial resolution multispectral (HSRM) image investigated in two ways. First, multispectral data is classified based on conventional approaches. Second, a method based on gray level co occurrence matrix (GLCM) as spatial objects extraction of the multispectral data is proposed. In this view, various texture parameters of the co-occurrence matrix method were used to highlight and extract the textures in the image. The method was computed on increasing matrix window size starting from original one. The Resourcesat-II Linear Imaging Self Scanning (LISS-IV) sensor multispectral image was used for testing the algorithms of the study area Phulambri Tehsil of Aurangabad region of Maharashtra state, India. The proposed approach was used as an input for Maximum Likelihood Classifier (MLC) and Support Vector Machine (SVM) approaches for identification and distribution of surface soil types and other patterns. The experimental outcomes of the present research were appraised on the basis of classification accuracy of methods. The overall accuracy of classification by MLC and SVM after spatial feature extraction was 92.82 and 97.32% with kappa value of 0.90 and 0.96 respectively. It was found that, the accuracy of the classification has increased after considering spatial features based on co-occurrence matrix. The results were promising to extract the mixed features for classification of soil type objects.