Multisensor Data Fusion for Coastal Boundary Detection by Res-U-Net Implementation Using High-Resolution UAV Imagery

Environmental planning, hazard monitoring, and coastal management depend critically on accurate shoreline definition. This work utilizes high-resolution unmanned aerial vehicle (UAV) data to develop a deep learning framework based on a Residual U-Net architecture for shoreline semantic segmentation. The proposed model integrates residual learning blocks into the conventional U-Net architecture to enhance gradient flow, improve feature extraction, and preserve fine boundary details in challenging coastal settings. Under a supervised learning framework, the model has been trained and validated using a dataset including UAV-acquired photos and manually annotated shoreline masks. The preprocessed input data has been reinforced by geometric adjustments and contrast normalizing to improve resilience and generalization. The Adam optimizer and binary cross-entropy loss helped the model be trained across 150 epochs. F1-score and Intersection over Union (IoU) measures have been used in quantitative performance evaluation. With a peak validation F1-score of 0.9483 and an IoU of 0.9018, the findings demonstrate that the Residual U-Net achieves great segmentation accuracy, showing robust spatial alignment with ground truth annotations. Visual analysis of the expected masks confirmed the approach's applicability to realworld situations by revealing consistent coastline localization throughout diverse environmental circumstances. This work presents a scalable and accurate method for operational shoreline monitoring, demonstrating the potential of deep residual structures for coastal boundary mapping using UAV platforms. Large-scale geospatial analytics and real-time coastal change detection can both benefit from the framework's extension to multi-temporal and multi-sensor data.