Empirical investigation of soft computing techniques in classification of remote sensed images

Abstract

In this dissertation pattern recognition in remote sensed images is examined Initially classification using Neural Network classifier and statistical methods are discussed The classification results are combined using decision fusion technique Following this clustering algorithms are discussed for optimizing the cluster centers in images The proposed methods in this thesis are validated using Remote sensed images A new neural network classifier is proposed which is unsupervised and works well in terms of classification accuracies The decision of this classifier is combined with the results of maximum likelihood classifier and minimum distance to means classifier using Boolean logic This improves the classification accuracy Support vector machines also provide a good solution for the classification problem It finds an optimal linear hyperplane by which expected classification error for the test samples is minimized This support vector classifier when amalgamated with maximum likelihood classifier using majority voting technique produced 92 classification accuracy