A Comparative Study of Transfer Learning, Convolutional Neural Network, and Random Forest for Satellite-Image-Based Land Use Classification

As the accessibility of satellite imagery grows, the necessity for precise classification models intensifies. This research endeavors to construct robust classification models capable of accurately discerning ten distinct land-use categories from satellite images. To attain this objective, the study delves into various methodologies, such as transfer learning, convolutional neural networks (CNNs), and the application of a Random Forest classifier in conjunction with Principal Component Analysis (PCA). Each approach presents distinct strengths and obstacles in tackling the intricacies inherent in satellite image classification. The dataset utilized for this investigation originates from the EuroSAT dataset, providing a standardized foundation for analysis and comparison. Through rigorous evaluation and comparison of these methodologies, this study aims to contribute to the advancement of land-use classification techniques, facilitating more accurate and efficient utilization of satellite imagery in diverse applications.