Remote Sensing Image Classification Methods Based on CNN: Challenge and Trends

Abstract

Remote sensing image classification occupies a vital place in earth observation and has many applications in military and civil fields. It can be divided into two typical tasks: high-resolution remote sensing images and hyperspectral image classification. However, high-resolution remote sensing and hyperspectral image classification cannot facilitate all features and achieve good accuracy with traditional methods. As deep learning methods, especially the convolutional neural networks (CNN), are developing rapidly, image classification methods based on CNN can perform well and provide new ideas for remote sensing classification. In this paper, we first review the background of typical remote sensing images and CNN. Then, we provide an overview of the development of the CNN model. After that, we point out some existing problems that we need to overcome for the CNN methods. Finally, the corresponding solutions are provided, and future work is presented with the analysis of some popular methods.