A Novel YOLOv3-tiny Network for Unmanned Airship Obstacle Detection

Abstract

Obstacle detection is an important issue in the study of an unmanned airship, which helps the airship to avoid obstacles and reduces the risk of accidences. This paper establishes an obstacle detection network, which is obtained by inserting wisely some $1\times 1$ and $3\times 3$ convolutional layers at the beginning and the end of the YOLOv3-tiny network. The experimental results show that our novel network leads to a higher accuracy compared to YOLOv3-tiny while with a satisfied processing speed.