<![CDATA[Coastal regions are among the most vulnerable ecosystems due to the combined impacts of natural processes and human activities. Climate change, population growth, and coastal development accelerate shoreline dynamics, increasing the need for accurate and efficient coastal monitoring. Satellite-based remote sensing, combined with deep learning techniques, provides a promising solution for large-scale and continuous shoreline analysis. This study proposes a deep learning–based approach for coastal land–sea segmentation using the ResNet50 architecture applied to Landsat 9 OLI imagery of the North Coast of East Java, Indonesia. The dataset consists of multispectral images processed into 224×224 pixel tiles, accompanied by manually generated ground truth segmentation maps. Two optimization strategies, Adam and Stochastic Gradient Descent (SGD), are evaluated to determine the most effective optimizer for improving segmentation performance. Experimental results demonstrate that the Adam optimizer outperforms SGD across multiple training epochs, achieving the highest segmentation accuracy with mean Intersection over Union (IoU) and Dice coefficient values of 0.888 and 0.934, respectively. These findings indicate that optimizer selection significantly influences the performance of ResNet50-based coastal segmentation. The proposed approach shows strong potential for supporting automated and large-scale coastal monitoring applications using medium-resolution satellite imagery.]]>
Copyrights © 2026
