Coastal land-use mapping along Johore Straits using Sentinel 1-SAR data with maximizing parameterization of machine learning classification

Abstract

In tropical regions, cloud cover accounts for a major obstacle to detect the coastal land-use while adopting remote sensing technology. The advent of the latest Sentinel-1 C-band synthetic aperture radar (SAR) satellite provides advantages to collect data in all-weathered conditions. In this study, Sentinel-1 images are processed using Lee filter and GLCM-Mean texture analysis in order to enhance the classification results. Several sets of parameters have been tested and these resulted in the optimum overall accuracy by Neural Network with 79.00% in 2015 and 68.29% in 2019. In contrast, Support Vector Machine classifiers obtained overall accuracies of 77.44% and 71.26% in 2015 and 2019 respectively. The results were accessed and it is found that Support Vector Machine outperforms the Neural Network classifier in discriminating data with high heterogeneity properties. Besides, Support Vector Machine has more consistent results in parameter testing compared to Neural Network.