Unconstrained ear recognition using residual learning and attention mechanisms