Deep learning in different remote sensing image categories and applications: status and prospects

Abstract

In recent years, the combination of deep learning and remote sensing has been a boiling state. However, because of the difference between remote sensing images and natural images, it is still an open question of whether deep learning methods can revolutionize remote sensing field. This work provides a brief review of 2674 related papers in deep learning with remote sensing (DL-RS) from 2014 to 2020. Keywords, publication years, journals, countries, and other essential characteristics of the papers were extracted. Also, we had set up some data items for information collection, such as remote sensing image categories, remote sensing applications, commonly used public datasets, and basic deep learning models. Our analysis shows that the number of research articles in DL-RS is still exploding, growing exponentially every year, and that DL methods have been applied to virtually all types of images and all applications of remote sensing. CNNs continue to be the most used deep learning model, accounting for 70% of all articles, GANs has now turned out to be the most used model after CNNs. Finally, we make some recommendations for future studies for the development of the DL-RS field.