Automated Oil Spill Detection and Segmentation in Marine Ecosystems Using Drone-Captured Images and Attention-Enhanced Convolutional Architectures

The primary objective of this project is to develop and evaluate novel deep learning architectures for automatic oil spill segmentation from drone?captured RGB imagery. The system aims to classify each pixel into three classes: Oil, Water, and Other (ships, sky, quays). Specific goals include: (1) addressing challenges such as ambiguous oil?water boundaries, irregular slick shapes, domain shift between port and open?sea datasets, and severe class imbalance; (2) designing lightweight yet robust models (HFC?Net and SSM?FusionNet) that integrate feature calibration, contour supervision, and spatial?spectral fusion.