Bridging the Accuracy Gap Between SNNs and DNNs via the Use of Pre-Processing for Radar Applications

Spiking Neural Networks (SNNs) are currently being explored as efficient solutions for performing AI tasks at the extreme edge. To fully exploit their potential, SNNs coupled to adequate pre-processing must be investigated. Within this context, a 4-b-weight SNN for radar gesture recognition is demonstrated, achieving a state-of-the-art 93% accuracy within only four processing time steps while using only one convolutional layer and two fully connected layers. In addition, the importance of signal pre-processing for achieving this high recognition accuracy in SNNs compared to deep neural networks (DNNs) is demonstrated with the same network topology and training strategy. It is shown that efficient pre-processing prior to the neural network is drastically more important for SNNs compared to DNNs. It is also demonstrated, for the first time, that the pre-processing parameters can affect SNNs and DNNs in antagonistic ways, prohibiting the generalization of conclusions drawn from DNN design to SNNs.