Deep Learning-Based Satellite Image Classification Using CNN and MobileNetV2 Transfer Learning
Authors: Divya.B, Subashini T.S, Bharathidasan.B
DOI: 10.87349/ahuri/181039
Page No: 111-125
Abstract
Classification of satellite images is a crucial task for remote sensing applications including land cover mapping, urban planning, and environmental monitoring. This work introduces a robust deep learning architecture for classifying high-resolution satellite images into four classes: cultivated area, forest area, desert region, and water bodies. The method uses both a native Convolutional Neural Network (CNN) and transfer fine-tuned MobileNetV2 model learning. Models were tested with accuracy, precision, recall and F1-score. The results demonstrate that MobileNetV2 achieved a test accuracy of 99.14%, which outperformed the custom CNN, which had attained 93.12%, thus confirming the efficacy of transfer learning in satellite image classification.
