BRAIN-INSPIRED SPIKING NEURAL NETWORKS FOR WI-FI BASED HUMAN ACTIVITY RECOGNITION

Human activities can be recognized through reflections of wireless signals which solve the problem of privacy concerns and restriction of the application environment in vision-based recognition. Spiking Neural Networks (SNNs) for human activity recognition (HAR) using Wi-Fi signals have been proposed in this work. SNNs are inspired by information processing in biology and processed in a massively parallel fashion. The proposed method reduces processing resources while still maintaining accuracy through using frail but robust to noise spiking signals information transfer. The performance of HAR by SNNs is compared with other machine learning (ML) networks, such as LSTM, Bi-LSTM and GRU models. Significant reduction in memory usage while still having accuracy that is on a par with other ML networks has been observed. More than 70% saving in memory usage has been achieved in SNNs compared with the other existing ML networks, making SNNs a potential solution for edge computing in industrial revolution 4.0.