A Few-Shot Learning Framework for Air Vehicle Detection by Similarity Embedding

Air vehicles such as aircrafts and drones have played an important role in surveillance and transportation for both civil and military applications. In this paper, we proposed a few-shot learning framework for air vehicle detection by similarity embedding, with a single moving camera mounted on another flying object. Firstly, we presented the example embedding with similarity conditioned LSTM-model for air vehicle detection. Secondly, we described the support set embedding with bidirectional LSTM-model of air vehicle training samples. Thirdly, we introduced the label prediction for air vehicle image blocks by attention kernel. Finally, we applied the fully convolutional network to segment air vehicle in the accurate bounding box. Experiment results of air vehicle detection show the effectiveness of our approach.