Deploying Emotion Recognition on Resource Constrained Device: Utilizing Light Weight Deep Learning Models with Multi-Physiological Signals

In recent years, the field of emotion recognition has witnessed significant advancements, with applications spanning from healthcare monitoring to human-computer interaction. However, deploying such systems on resource-constrained end devices, such as wearable sensors or mobile phones, remains a challenging task due to limitations in computational power, memory, and battery life. This paper proposes a novel approach to address these challenges by utilizing lightweight deep learning models that are specifically designed to operate efficiently on such devices. The proposed system integrates multi-physiological signals, including Electrocardiogram (ECG), Photoplethysmographic (PPG) readings, Heartbeat Rate (HR), Skin Temperature measurements (SKT), and Respiratory patterns (RSP) to provide a comprehensive assessment of emotional states. We evaluate the performance of our system on a dataset of physiological signals collected from subjects undergoing video emotional stimuli. Experimental results demonstrate that our lightweight models achieve comparable accuracy to heavier models while significantly reducing computational overhead and power consumption. The findings of this study pave the way for the development of practical, real-time emotion recognition systems that can be deployed on a wide range of resource-constrained end devices.