U-SPDNet: An SPD manifold learning-based neural network for visual classification

Rui Wang, Xiaojun Wu, Tianyang Xu, Cong Hu, Josef Kittler

Abstract: Highlights•This paper designs a U-shaped neural network in the context of SPD manifolds (USPDNet) to cope with the problem of statistical information degradation.•The geometric computation-based skip connections are added from the encoder to the decoder to improve the representational capacity of the proposed SPD network.•Extensive experiments show that our method can achieve improved accuracy, even with limited data.