Global warming, storm surges, and coastal erosion damage the environment. Complex coastal environments with dynamic features make manual digitisation and threshold-based segmentation inefficient and inaccurate. These methods are error-prone and hard to scale because they use predefined thresholds and human input. For coastal monitoring and management, convolutional neural networks (CNNs)-based deep learning models like UNet and DexiNed offer more accurate, scalable, and automated coastline extraction and erosion analysis. UNet and DexiNed were compared to traditional coastline extraction and erosion analysis methods for accuracy, precision, recall, and F1-score. The models were tested for environmental monitoring and coastal zone management, especially in climate-vulnerable regions. Overcoming manual input and threshold-based segmentation, satellite-trained UNet and DexiNed models accurately identified coastline boundaries in complex coastal environments. Accuracy, precision, recall, and F1-score assessed performance. GIS showed Bangkok Bay coastal erosion from sea-level rise and storm surges. The 2010–2024 dataset included Google Earth, Sentinel-2, and Landsat coastal imagery of Bangkok Bay. Both UNet and DexiNed outperformed traditional methods in all evaluation metrics. DexiNed detected edges, especially land-water boundaries, better than UNet at pixel-wise segmentation and coastline detection. Climate change is affecting vulnerable regions like East Bangkok Bay and South Coastal Areas, as GIS-based erosion analysis showed significant coastline shifts. This study detects coastlines precisely, scalable, and automatically using AI-driven deep learning models UNet and DexiNed. The method is ideal for Coastal and Water Resource Engineering, providing robust dynamic shoreline monitoring, erosion prediction, and climate resilience planning.