Evaluation of convnets for large-scale scene classification from high-resolution remote sensing images

Convolutional neural networks (convnets) have made possible a number of breakthroughs in image classification and other computer vision problems. However, in order to successfully apply convnets to a new task it should be trained on a large set of labeled samples. Acquisition of a large number of manually labeled remote sensing images requires highly trained analysts which makes it a very expensive task. This is the main reason why we still lack large training sets of remote sensing images. Nevertheless, convnets can still be applied to remote sensing image classification by virtue of using convnets pretrained on another large dataset and fine-tuned to the task at hand. In this paper we investigate the use of pretrained and fine-tuned convnets for both end-to-end classification and feature extraction from remote sensing images. We analyze the quality of the features extracted from various layers of the network from the standpoint of classification accuracy. Using a fine-tuned ResNet we obtain classification accuracy of over 94% on challenging AID dataset.