 In this paper, we explore the potential of using pre-trained convolutional neural networks, CNNs, for high-resolution remote sensing, HRS, scene classification. We propose two scenarios for generating image features from these CNNs, one involving extracting activations from fully connected layers and another involving dense features from the last convolutional layer at multiple scales. Experiments show that both scenarios outperform traditional low and mid-level features and that combining features from different CNNs leads to further improvement. This suggests that pre-trained CNNs can be used to generate effective image features for HRS scene classification. This article was authored by Fan Hu, Guesong Xia, Jingwen Hu, and others.