Remote Sensing Classification Based on Improve Ant-Miner Algorithm: A Case Study of Alwar, Rajasthan, India

Abstract

Earth Observation Satellite (EOS) image itself contains image ambiguity. Various conventional methods like minimum distance to mean or maximum likelihood image clustering algorithm do not meet the accuracy that have required by user in the virtue of cost-effective land use/land cover classification. In Ant colony optimization (ACO), association rule mining is a prevalent and well researched method for discovering useful relations between variables in large databases. The proposed work presents an Ant-Miner technique that can be adapted according to the database of expert knowledge for a more focused satellite image classification. This paper investigates the principle of traditional rule mining, which will produce more nonessential candidate sets when it reads data into candidate items. Particularly when it deals with massive data like land use/land cover training sets data, if the minimum support and minimum confidence are relatively small, combinatorial explosion of frequent item sets will occur and computing power and storage space required are likely to exceed the limits of computer. A conventional Ant-Miner algorithm is proposed and used in rules mining. The experiment results showed that execution time of proposed algorithm is lower than traditional algorithm and has better execution time and accuracy.
Keywords: Ant colony optimization, remote sensing, LISS, MATLAB, error matrix