Training Dataset for Sandy Beach Detection Derived from IRS ResourceSat-2/2A LISS-IV Imagery Using a U-Net Model

This dataset contains training, validation, and test image tiles prepared for sandy beach segmentation along the Indian coastline using a U-Net deep learning model. Fourteen representative coastal sites were selected to capture the diversity of geomorphic settings, sediment compositions, and shoreline morphologies. Manual annotation was performed on high-resolution IRS ResourceSat-2/2A LISS-IV imagery to delineate sandy beach extents, ensuring accurate ground-truth masks. In addition to the original spectral bands (green, red, and near-infrared), derived indices such as the Normalized Difference Vegetation Index (NDVI), Green–NIR ratio, and a composite intensity measure were included as additional input channels to enhance feature separability between sand, vegetation, and water. All imagery and masks were partitioned into 128 × 128 pixel tiles (~742 × 742 m) with 40% overlap to ensure complete coverage and preserve shoreline continuity. The dataset is divided into training (60%), validation (20%), and test (20%) subsets comprising a total of 1,088 samples. It supports reproducible research in coastal geomorphology, shoreline mapping, and automated sandy beach detection using deep learning approaches.