A genetic algorithm rule-based approach for land-cover classification

Classification of land-cover information using remotely-sensed imagery is a challenging topic due to the complexity of landscapes and the spatial and spectral resolution of the images being used. Early studies of land-cover classification used statistical methods such as the maximum likelihood classifier. Recently, however, numerous studies have applied artificial intelligence techniques – for example, expert system, artificial neural networks and support vector machines – as alternatives to remotely-sensed image classification applications. There is a major drawback in applying these models that the user cannot readily realize the final rules. In this paper, a rule-based classifier derived from improved genetic algorithm approach is proposed to determine the knowledge rules for land-cover classification done automatically from remote sensing image datasets. The proposed algorithm is demonstrated for two image datasets classification problems. Results are compared to other approaches in the literatures. The preliminary results indicate that the proposed GA rule-based approach for land-cover classification is promising.