Boosting Artificial Intelligence Performance in Sentinel-2 Imagery Analysis

Abstract

The combination of remote sensing with AI methods such as deep learning, along with tools like geographic information systems, has transformed both spatial analysis and how decisions are made in a variety of fields. In this chapter, the authors show how important these AI technologies and techniques can be for improving the accuracy of information produced by remote sensors when it comes to monitoring water resources managing them sustainably, assessing flooding risks and environmental monitoring, thus contributing to a more informed and sustainable future. Accurate identification of water bodies using satellite images is crucial for purposes such as managing water resources, the environment, urban planning, and reacting to disasters. This study evaluates the performance of three advanced deep learning models DeepLabV3+, U-Net, and FCN in the semantic segmentation when looking at water bodies in pictures taken by Sentinel-2 satellites. The three models were assessed by metrics of accuracy, recall, precision, F1score, and Dice coefficient. The best performer was U-Net: it scored highest overall.