Oriented Feature Augmentation for Few-Shot Object Detection in Remote Sensing Images
Zhenwei Zhang, Jianguo Hao, Chongyu Pan, Guang Ji · 2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI) · 2021
In recent years, many researches have been conducted for few-shot object detection in image of natural scenery. But in the field of remote sensing images, this is a more challenging task. due to the fact that the objects have arbitrary direction in remote sensing image compared with the natural scenery images, which will lead to larger intra-class distance, and are more likely to cause category confusion. Toward this end, we proposed a method of oriented feature augmentation (OAF) on the basis of dual pipelines solution. The method is simple without bells and whistles. With the help of enhanced category features, the detector is easier to achieve classification and regression of remote sensing objects. Extensive experiments on NWPU VHR-10(v2) and DIOR dataset in various settings show that, our method achieves new state-of-the-art performance by a large margin, demonstrating its effectiveness and generalization ability.