A multi-objective cooperative coevolutionary approach for remote sensing image classification

A current problem that classification algorithms on high-resolution satellite images are addressing is the feature selection problem (FSP). The number of object's features on satellite images is often very large. However, not all these features contribute equally to the classification results. There are still many redundant, unrelated features. It is, therefore, necessary to select these features before performing the classification step. Besides, with the selected feature set, selecting a suitable classifier also plays a very important role. In this study, the authors solve this problem with a multi-objective co-operative co-evolutionary approach (MCCA). In the MCCA, we use two populations evolving together: one population helps to find the most important set of features (named Feature population) and the other helps to get the most appropriate classifier (named Classifier population). The performance of the MCCA is examined on three satellite image datasets. From experimental results, the proposed algorithm has shown the efficiency in improving classification accuracy as well as reducing the number of characteristics. © 2019 IEEE.