Deep Transfer Learning with RESNET for Remote Sensing Scene Classification

This paper explains a comparative examination of aerial images for Land-Use and Land-Cover (LU&LC) classification, focusing on urban regions where scattering properties are nearly equivalent. To attain better performance metrics, a complete analysis was carried out, in which two ResNet-based models were applied to the RSI-CB128 dataset utilizing three different optimizers: Adam, SGDM, and RMSProp. Both ResNet models have been used on five different classes of the dataset with three different optimisers. RMSprop has the best accuracy with moderate simulation time in the case of ResNetl8, while in the case of ResNet50, SGDM has the best accuracy of 95.8% with the lowest simulation time.