SSP: A Large-Scale Semi-Real Dataset for Semantic Segmentation of Spacecraft Payloads

Semantic segmentation technology is widely used in autonomous driving, environment perception, and other fields. With the development of deep learning technology, semantic segmentation will also be applied to spacecraft-related research (i.e., spacecraft-payload segmentation). However, the lack of spacecraft-related datasets seriously hinders the development of deep learning technology in research fields, such as spacecraft control and payload segmentation. To address this issue, we release a novel semi-real dataset for the segmentation of spacecraft payload. Specifically, the images in our released dataset are generated by merging images of spacecraft models taken by a real camera and space images released by the National Aeronautics and Space Administration (NASA). Furthermore, we propose a network SPSNet with an anti-pyramid-structured decoder for the segmentation of spacecraft payload. Experimental results show that our proposed network achieves superior performance compared with the well-known networks on our released dataset.