An Equivariant Neural Network with Hyperbolic Embedding for Robust Doppler Signal Classification

This paper focuses on the robustness aspects of Doppler signal processing tasks with Machine Learning algorithms within the context of pathological radar clutter classification. More precisely, group-based convolution operators are combined with the hyperbolic embedding technique to build an equivariant neural network operating on the Doppler signal represented as complex covariance matrices. Our numerical testing performed on simulated data has shown the superiority of our approach when compared to conventional neural networks, from both accuracy and robustness standpoints.