Attention-Based ResNet for Accurate Water Body Classification Using Spectral Indices

Accurate classification of water bodies, such as rivers, lakes, and seas, is crucial for effective environmental monitoring and resource management. This research integrates both spectral water indices and deep learning models to improve water body classification using high-resolution satellite imagery from the EuroSAT dataset. Initially, water index methods were applied, followed by deep learning techniques using ResNet50 and ResN et152 architectures. To further enhance model performance, attention mechanisms were proposed and incorporated into the ResN et models, enabling the network to focus on key features. Among the tested models, ResN et50 with attention layers achieved the highest accuracy at 86 %. These findings highlight the effectiveness of attention mechanisms in improving classification accuracy, providing valuable insights for future applications in remote sensing and geographic information systems.