A Classification Method for Satellite Imagery Utilizing Transfer Learning Techniques

This study focuses on land cover classification by determining the category of remote sensing images according to specific target class labels. The objective involves categorizing high-resolution satellite images from the EuroSAT dataset into ten distinct classes, each representing a markedly different type of land cover. To enhance the precision and effectiveness of the classification process, transfer learning was utilized to refine five prominent pre-trained models: VGG16, VGG19, ResNet50, MobileNetV2, and ResNet101V2. Our analysis of the experimental data indicates that ResNet50 provided the best results based on the models’ classification accuracy among the models assessed, with a mean accuracy of 98.07%. This finding further establishes the potential of employing deep learning techniques, particularly transfer learning to tackle complex challenges in remote sensing. The suggested methodology may also reflect the increasing significance of artificial intelligence in geographic and environmental endeavors, with extensive applications in environmental monitoring, land use analysis in spatial planning, and urban development.