EAGLE: An Enhanced Attention-Based Strategy by Generating Answers from Learning Questions to a Remote Sensing Image

Yeyang Zhou, Yixin Chen, Yimin Chen, Shunlong Ye, Mingxin Guo, Ziqi Sha, Heyu Wei, Yanhui Gu, Junsheng Zhou, Weiguang Qu

Abstract: Image understanding is an essential research issue for many applications, such as text-to-image retrieval, Visual Question Answering, visual dialog and so forth. In all these applications, how to comprehend images through queries has been a challenge. Most studies concentrate on general images and have obtained desired results, but not for some specific real applications, e.g., remote sensing. To tackle this issue, in this paper, we propose an enhanced attention-based approach, entitled EAGLE, which seamlessly integrates the property of aerial images and NLP. In addition, we contribute the first large-scale remote sensing question answering corpus (https://github.com/rsqa2018/corpus). Extensive experiments conducted on real data demonstrate that EAGLE outperforms the-state-of-the-art approaches.