Abstract
The process of classifying images involves grouping pixels with similar characteristics into one class or cluster. Traditional pixel-based classification methods such as support vector machine, random forest, and decision tree yield poor results for synthetic aperture radar imagery (SAR) because of limited spectral information. This chapter provides the results of a two-dimensional (2D)-convolutional neural network (CNN)-based classification using Sentinel-1 (SAR) data over the Central State Farm, Hissar, Haryana, India. Hyperparameters of 2D-CNN were optimized using Bayesian optimization. Several textural features derived from S1 data were also layer-stacked with both vertical–vertical (VV) and vertical–horizontal (VH) polarized images. Results suggest that using texture features obtained from both VV and VH images improved classification accuracy for the considered area.
