Automating shoreline extraction from beach imaging systems for coastal monitoring

High-frequency coastal monitoring is essential for understanding shoreline dynamics under climate change and supporting coastal adaptation. Although ground-based remote sensing offers a scalable solution for data acquisition, automated shoreline extraction from beach imaging systems remains challenging due to variable image quality, complex geometries, and environmental conditions. This study presents a robust deep learning framework for instantaneous shoreline extraction from crowd-sourced imagery (i.e., smartphone captures). We benchmarked state-of-the-art segmentation models against a traditional histogram-based deterministic method. Results show that convolutional neural network approaches, particularly DeepLabV3 with 512 ×512 patches, significantly outperform conventional methods, achieving spatial reliability comparable to manual digitisation. Importantly, the model overcomes perspective distortion and interpolation artefacts in rectified imagery across diverse tidal environments, showing strong environmental robustness despite site diversity and data imbalance. The framework has been deployed as an open-source, containerised module within the European Open Science Cloud, enabling transferable and operational image-based coastline monitoring.