Comparative Analysis of Modern Methods for Surface Type Identification in RGB Image Data

Abstract

The rapid development of drone technology and deep learning algorithms also expands the possibilities of environmental monitoring, e.g., the search and management of water bodies. This study aims to harness these advances for the accurate identification of surface types, with a specific focus on water bodies, in RGB image data. Using a dataset comprised of aerial images captured over the Baroch Pond within a nature reserve in the Czech Republic, this study comparative to evaluate the performance of deep learning models, including U-Net, Pyramid Scene Parsing Network (PSPNet), and DeepLabV3, in classifying surface types. The classification accuracy is slightly over 90% for most deep learning algorithms. These results show the potential of deep learning in this area. And this is key for a number of interested parties, for example, state administration, water resource managers, farmers, and tourism industry.