Akademik Veri Yönetim Sistemi
INTERNATIONAL GLOBAL SUSTAINABILITY AND DEVELOPMENT CONGRESS, Kayseri, Türkiye, 15 - 16 Ekim 2025, ss.80-87, (Tam Metin Bildiri)
Remote sensing is a crucial tool for ensuring the sustainable use of the world's natural resources. It provideslarge-scale, continuous and objective data for sustainability studies. In recent years, it has been frequently usedin conjunction with deep learning-based models to infer water areas. Semantic water segmentation, performedto identify water areas, enables the identification and tracking of water resources. This process creates a crucialdata infrastructure for sustainable water management, ecosystem protection, and drought mitigation. In thisstudy, the ERCNET network model was developed using WATERNET to enable the automatic inference ofwater bodies. The model has fewer layers and fewer filters compared to complex deep learning models suchas FC-DenseNet, Pix2Pix, Dilated U-Net, Standard U-Net, and Fractal U-Net, which are commonly used inimage segmentation. Therefore, it can be described as a simpler, shallower version of the widely used U-NETarchitecture. This shallow structure of the model facilitates its applicability to small-scale datasets. Therefore,it is also compatible with sustainable artificial intelligence applications because it uses less computationalpower and consumes less energy. According to the results obtained from the study, the model's accuracy wascalculated as 99.77%, Recall as 99.41%, F1 as 99.52%, Precision as 99.63%, and IOU as 99.06%. These resultsdemonstrate that ERCNET can achieve high accuracy despite its low number of parameters. It alsodemonstrates that it offers a more efficient solution compared to other methods.