Adaptive Coastline Detection based on Deep Q-Network-Guided Contour Refinement

TL;DR

A novel framework combining Constrained Deep Q-Learning with Precise Sea Area Cleanup that introduces three key innovations: HSV-based dual-stream feature extraction for robust water-land discrimination, context-aware action constraints with curiosity-driven exploration to prevent artifacts and a single-image reinforcement learning method that adapts to new environments without retraining.