High-Fidelity Emotion Recognition via SDR-Based Wireless Sensing and Deep Learning

Emotions pertain to the internal and personal perceptions and reactions people exhibit in response to various stimuli or circumstances. Human emotion recognition is a rapidly developing field that aims to detect and understand emotional states using cues such as facial expressions, speech, gestures, and physiological signals. Unlike traditional non-contact human emotion recognition techniques, this paper introduces an innovative method that utilizes radio frequency (RF) technology to detect and analyze human gestures and activities for the associated emotion recognition using deep learning algorithms. We propose a non-contact software-defined radio (SDR) based RF and time-series-based deep learning framework for human emotion recognition through activity sensing. An orthogonal frequency division multiplexing (OFDM) transceiver is designed to measure human activity and gesture imprints based on wireless channel state information (WCSI) in a controlled laboratory environment with data collected from single individual at a time. We use sequential deep learning models, simple recurrent neural networks (RNN), long short-term memory (LSTM), bi-directional LSTM (Bi-LSTM), gated recurrent units (GRU), and bi-directional GRU (Bi-GRU) to classify eight human emotions namely anger, disgust, fear, happiness, interest, sadness, shame, and surprise. The performance of these models is evaluated based on the accuracy, precision, recall, and F1 score metrics and achieved classification accuracies of 99.61%, 99.68%, 99.71%, 99.66%, and 99.78%, respectively.