ABSTRACT
Chapter 4 primarily focuses on the implementation of various state-of-the-art machine learning algorithms for land use and land cover classification on multifrequency band Synthetic Aperture Radar (SAR) dataset. It plays a major role in modern urban planning, management, and sustainable expansion. It uses the spaceborne multifrequency SAR data captured by respective sensors and the study area of San Francisco, California, USA, for detailed mapping of man-made and natural objects. It implemented various machine learning models like k-Nearest Neighbor, Random Forest, Support Vector Machine, and Naïve Bayes classification models on ALOS-2/PALSAR-2, GF-3, RADARSAT-2, and TerraSAR-X dataset. Each sensor and frequency band has a different capacity to identify the target with modest accuracy. This study will also be of help to the research community in selecting a model per the available dataset to produce better classification results. The Support Vector Machine classification model is found to perform better among all models for the respective dataset with an overall pixel accuracy of 94.74% and an F1-score of 0.95 on the TerraSAR-X dataset.
