Implementasi Deep Learning untuk Pelabelan Lahan pada Citra Satelit Area Desa Leles

Abstract

Efficient land management in rural areas is essential for environmental sustainability and community well-being. This study develops a deep learning model to classify land types in satellite imagery, focusing on two main categories: residential areas and vacant land. The dataset consists of 2,080 cropped images from Google Earth, divided into 790 images for training, 790 for validation, and 500 for testing. The model is trained using transfer learning with the ResNet50 architecture and optimized over 50 epochs. Training results show an accuracy of 99% for training and 98% for validation, while evaluation on the test dataset achieves an accuracy of 91.4%. The model successfully identifies most images correctly, although 43 out of 500 images were misclassified. These findings demonstrate the significant potential of deep learning in supporting more accurate land management based on satellite imagery, contributing to more efficient spatial planning.