Few-Shot Object Detection of Drones
Zou Weibao, Xindi Liu, Yang Jitao, Wei Qu · 2021 International Conference on Electrical, Computer and Energy Technologies (ICECET) · 2021
Nowadays, object detection has experienced great progress, especially in small target detection. And the drone has the characteristics of small size, various shapes, and flexible flight. So how to detect the drone based on deep learning is a topic worthy of study. This paper aims to build sample data sets of the drone by means of the data augmentation and improve the detection accuracy of the small drone based on the You Only Look Once v5s (YOLOv5s). This model is adopted in the study since it can achieve a good trade-off between accuracy and speed. In the paper, transfer learning is used to solve the problem caused by small data set. 245 Images which are trained and tested in this study are from the Internet. The experimental results show this proposed method can achieve 96.8% mean average precision(mPA50) and accurately detect the moving small drone in the video.