RANC-Based Hardware Implementation of Spiking Neural Network for Sleeping Posture Classification

Numerous studies have proven that sleeping position classification plays an essential role in medical diagnosis. Currently, spiking neural networks (SNNs) are emerging as a new trend to solve this task due to its energy-saving advantage. However, none of current studies considers the hardware implementation of their proposed networks. Therefore, this paper presents a process to synthesize a spiking neural network for sleeping posture classification using RANC on FPGA. To conserve hardware resources, we proposed a method called parameter reconfiguration, which could reduce the number of SNNs model cores required from 84 to 21. Experimental results confirmed that our hardware implementation achieved a high accuracy of 92.4 $$\%$$ and low-power consumption of 0.0167 (W) per image.