A Lightweight Real-Time Semantic Segmentation Network for Equipment Images in Space Capsule

PubDate: October 2020

Teams: Space Engineering University；China Astronaut Research and Training Center

Writers: Zhongkai Ma; Jin Yang; Jiangang Chao; Wanhong Lin

PDF: A Lightweight Real-Time Semantic Segmentation Network for Equipment Images in Space Capsule

Abstract

The combination of semantic segmentation technology and augmented reality technology can provide auxiliary information when astronauts train in augmented reality mode, which will greatly improve the training efficiency and reduce mishandling for astronauts. However, the equipment in space capsule have the characteristics of irregular shape, similar texture and small target while the mixed reality application requires high real-time performance, the above factors bring challenges to the context consistency, accuracy and real-time of semantic segmentation. In response to the challenges, referring to [3], one of the best lightweight real-time segmentation networks, a new network is specially designed for our application. Experimental results show that the designed network can obtain competitive segmentation results on target dataset and better real-time performance than classic networks such as [3]. Overall, the designed network meets the requirement.