Evaluation of Transfer Learning Approach for Satellite Image Classification

In this paper, the usability of the Convolutional Neural Network (CNN) models built by the transfer learning to determine the land use from satellite images and the effect of transfer learning on the classification accuracy were investigated. Within the scope of this study, transfer learning approach has been evaluated by using different public datasets containing satellite images for academic studies. The CNN models fully trained and the CNN models trained using transfer learning were compared in aspect of accuracy to classify satellite images. It has been observed that the CNNs trained using transfer learning yielded more accurate results for each dataset used in the study.