Shoreline Change Analysis with Deep Learning Semantic Segmentation Using Remote Sensing and GIS Data

Abstract

Shoreline management is essential for navigation, coastal resource management, and coastal planning and development. Shoreline change detection is vital for shoreline monitoring; however, traditional methods used for such detection are laborious and have limited accuracy. An approach that integrates remote sensing imagery and geographic information systems (GISs) is proposed herein to simultaneously identify shoreline changes and perform grid-level visualization for updating shoreline data. The integrated approach uses deep learning-based segmentation networks and water indexes to accurately classify land and sea in remote sensing images. Transfer learning was used to address the issue of insufficient data, wherein weights trained on a large open dataset were applied to the target area. The segmentation results were compared with existing shoreline GIS data to identify the areas experiencing shoreline changes. Grid-level visualization enhanced the identification of regions requiring flexible data updates and investigation efficiency by focusing on specific areas. The proposed approach accurately detected shoreline changes, albeit with some errors of commission, predominantly in regions featuring intricate shorelines and small clusters of islands. The proposed approach offers efficient solutions for shoreline change detection, with potential applications in coastal management, environmental science, urban planning, and coastal hazard assessment.