Comparative study of Neural Network approach and Genetic Programming approach to Land Cover Mapping

This paper explores the feasibility of applying Neural Networks and Genetic Programming to Land Cover Mapping problem. Land Cover Mapping has been done traditionally by using the Maximum Likelihood Classifier (MLC). Neural Networks (NN) and Genetic Programming (GP) classifiers have advantage over statistical methods because they are distribution free, i.e., no prior knowledge is needed about the statistical distribution of the data. Neural Network has been applied for the classification but we may not be sure of getting the optimal solution. GP has the ability to discover discriminant features for a class. GP has been applied for two-category(class) pattern classification. This idea is extended to -class image classification problem by modeling the problem into two-class problems, and a genetic programming classifier expression(GPCE) is evolved as a discriminant function for each class. The GPCE is trained to recognize the samples belonging to its own class and rejecting samples belonging to other classes. Experimental results are presented to demonstrate the applicability of Neural Network and GP for land cover mapping problem, and the results are found to be satisfactory.