Synergistic Exploration Combining Traditional And Evolutionary Methods To Improve Supervised Satellite Images Classification

Abstract

This paper aims to enhance the performance of supervised classification of satellite images by adopting a spectral classification approach, which often encounters the issue of class confusions due to its reliance solely on spectral informat ion. The proposed approach, EAMD (Evolutionary Algorithm and Minimum Distance), integrates a Genetic Algorithm-based evolutionary method with the Minimum Distance method. During the training phase, the Genetic Algorithm generates an optimal set of subcategories to represent different object classes present in the image and identifies an optimal representative set of pixels for class assignment. Experimental tests conducted on various satellite images yield promising results, demonstrating the capability of Genetic Algorithms to enhance classification accuracy and effectively identify and exclude misleading pixels responsible for class confusions. This aspect is crucial, as the effectiveness of supervised classification heavily depends on the quality of the training samples. Validation of the approach was further reinforced by intentionally injecting erroneous data into the training data. Compared to the Minimum Distance method, the proposed approach successfully detects and avoids the erroneous pixels, a task unaccomplished by the Minimum Distance method. The obtained results demonstrate that the hybrid proposed approach offers significant potential for improving the accuracy and reliability of satellite image classification techniques.