Developing a statistical learning based modified classifier for remote sensing purposes

The sicience of learning plays a key of role in the fields of statistics, data mining and artificial intelligence, intersecting areas of engineering and other disciplines. EM algorithm is an effective analysis technique in the field of statistical learning, a type of learning which is effectively used in many fields of science, finance and industry. It is also a popular tool for simplifying difficult maximum likelihood problems. The situations such as sensitivity to initial estimation values, local minima problem, faster convergence, and singularity of variance covariance matrices are certain problem leading to the consequences that EM algorithm cannot make parameter estimates or make inaccurate ones. In this dissertation, theoretical structure of EM algorithm was examined and a matlab code was developed accordingly. Later, this code was used to make parameter estimates for various classes available in remote sensing data and image classification field with the help of these parameter estimates. It was observed that variance covariance matrix or matrices had singularity trap problems; however, image classifications were successfully completed after EM algorithm was modified to solve these singularity trap problems. Finally, image classification results obtained by using several classical classifiers available in the field of remote sensing were compared with those obtained through unmodified and modified EM algorithm. It was observed that image classifications obtained through unmodified and modified EM algorithm were more reliable. The main objective of this dissertation is to develop a new statistical learning based modified classifier that will provide more reliable results than supervised and unsupervised and distance based classifiers that do not use probability distribution. It also deals with algorithms that will provide solutions to the problems mentioned above. Keywords : Remote Sensing, Statistical Learning, EM Algorithm, Modified EM Algorithm, Image Classification