A new optimization based approach to the unsupervised classification of satellite images

The classification process is one of the methods used for the analysis of remotely sensed images. The unsupervised classification methods do not require training data as opposed to supervised classification methods. Collecting quality and sufficient training data directly affecting classification accuracy is a laborious and costly process. This process requires the user to be familiar with the area of study and to become an expert in the classification methods. Unsupervised classification methods are more practical since they do not require training data. However, certain inputs need to be specified by the user, such as the number of classes, the maximum number of iterations, and the threshold valuesthat indicate when the classification procedure will be terminated. In this study, a new unsupervised classification method that eliminates user dependence and produces high accuracy results is proposed. This method consists of data extension, useful data selection, segmentation and optimization stages and performs the classification automatically. This developed method contributes to the literature in terms of achieving successful classification results which can be easily applied by people from different professional disciplines without being an expert in image processing. In addition, a new approach has been proposed to label unlabeled results by unsupervised classification methods. Key Words: Unsupervised Classification, Optimization, Remote Sensing, Land Cover Labelling