| Popis: |
Coastal erosion due to extreme events can cause significant damage to coastal communities and deplete beaches. Post-storm beach recovery is a crucial natural process that rebuilds coastal morphology and reintroduces eroded sediment to the subaerial beach. However, our understanding of this process is limited, which hinders effective management of coastal risks and vulnerability.In this study, we propose a new semantic segmentation method based on Convolutional Neural Networks (CNN) to detect changes in sand composition on the beach over time using high-temporal resolution video monitoring. We validate our model using a set of beach imagery recorded at Cedar Lakes, Texas, after Hurricane Harvey in 2017. We train and predict on image patches to minimize information loss from rescaling. Our CNN-based model achieves high accuracy (mean accuracy of 95.1\% and mean IOU of 86.7\%) and introduces a new metric to measure potential false positives and negatives.Our study also discusses how to identify blurry or rainy beach images in advance of semantic segmentation prediction, as our model is less effective in predicting these types of images. Our findings provide valuable insights into post-storm beach recovery and demonstrate the potential of high-frequency video monitoring and CNN-based semantic segmentation for improving our understanding of beach recovery processes. |
