Semi-automatic general approach to achieve the practical number of clusters for classification of remote sensing MS satellite images

• Published in: Spatial information research (Online) Vol. 28; no. 2; pp. 203 - 213
• Main Authors: Serwa, A., El-Semary, Hossam H.
• Format: Journal Article
• Language: English
• Published: Singapore Springer Singapore 01.04.2020; 대한공간정보학회
• Subjects: Database Management; Earth and Environmental Science; Geographical Information Systems/Cartography; Geography; Remote Sensing/Photogrammetry; 공학일반; K-means; Unsupervised classification; Multi spectral scanner (MSS); Remote sensing
• ISSN: 2366-3286; 2366-3294
• DOI: 10.1007/s41324-019-00283-z

The main objective of this research is to find a semi-automatic method to determine the practical number of clusters in MS satellite images. This study puts a general skeleton for determining the practical number of classes in multi spectral (MS) remote sensing images. The sequence of the research starts with input the reference data, proposed classes’ samples and the MS image of the study area. The unsupervised classification is carried out using Envi software many times with different excessive number of classes. Fuzzy K-means method is applied as an unsupervised classification algorithm. A comparison between the classified image and the proposed classes’ samples is carried out using ADIPRS software to testify if the classification is reliable or not based on appearance. The process continues until the condition of the appearance is satisfied then the comparison with the reference is carried out to test the accuracy limit of the classes.