Comparison of Satellite Images Classification Techniques using Landsat-8 Data for Land Cover Extraction

Accurate extraction of land cover types from thematic maps using satellite images still constitutes a critical challenge. The selection of a suitable satellite image classification algorithm is considered a crucial prerequisite for successful classification results that are required for various applications. The optimal classification algorithm is considered a significant key for improving classification accuracy. The principal foci of this study were to compare, analyze the performance, and assess the effectiveness of four classification algorithms including ISODATA, K-means, pixel-based and segment-based classification techniques to attain accurate land cover extraction from remote sensing data. The classified images were validated with ground control points obtained from field visits in addition to the DigitalGlobe and Google Earth Pro. The overall accuracy of the ISODATA, K-means, pixel, and segment-based classifications were 81.82%, 77.27%, 92.42%, and 87.88%, respectively. The results revealed that the pixel-based classification presented a superior in terms of the overall accuracy and kappa coefficient.