Supervised Classification of Multispectral Images: A Comparative Study

Abstract

This work focuses on a satellite data supervised classification which is one of the most important topic in satellite image processing domain. A comparative study is carried out between statistical and commonly used classifiers in one hand, and relatively new adaptive classifiers that fall under the umbrella term of metaheuristic classifiers, on the other hand. The general purpose of these classifiers is to generate the most accurate land use and land cover mapping. The comparative study aims to state of the metaheuristic classifiers (bio-inspired algorithms seen as optimization and minimization algorithms) robustness over the statistical classifiers. The main drivers for investigating these methods are also addressed. The concerned classifiers are used for a supervised classification of a remotely sensed multiband image covering the East part of the capital city Algiers. Following the resulted classification maps and their assessment, it is concluded that 1) statistical classifiers are not as effective and reliable as metaheuristic classifiers, and 2) the immune classifier may offer a viable alternative to the genetic classifier in terms of precision and time consuming.