Development of Evidence Classifier Using Hybridization Method for Improved Satellite Image Classification

摘要

In this research study an attempt is made to develop a new algorithm for land use/cover classification based on hybridization of existing supervised image classifiers namely parallelepiped and minimum distance to means. This algorithm is then applied on the area related to Sri Ram Sagar reservoir for its validity to generalize the concept. In this approach the computerized interpretation of pixels are considered in an iterative process of inferencing technique , The iterative process continue until one of the following two conditions are satisfied. These conditions are i) a state having a single class is reached: In such cases it was deemed that a firm decision has been reached.ⅱ) No new state is reached after an iteration: In such cases a firm decision cannot be reached. This algorithm is then applied on the data set of Sri Ram Sagar reservoir area, which is obtained from NRSA received by the IRS-1B satellite, LISS-II sensor. This algorithm named as Evidence Classifier is compared with the existing supervised image classifiers namely parallelepiped and minimum distance to means by using the same data set. The comparative study revealed that the evidence classifier yield better results, uses less quantity of training data and consumes less computation time than maximum likelihood and minimum distance to means classifiers and the classification accuracy is relatively high and more than 90%. It is recommended that the evidence classifier can be applied if the project area is connected with any reservoir such as Sri Ram Sagar reservoir of the present study.