Enhancing Wearable Motion Monitoring Devices: A Multi-level Attention Fusion Approach for Reliable Fatigue Assessment

Abstract

Abstract This study aims to create a cost-effective, reliable motion monitoring device that can thoroughly analyze a subject's fatigue. It achieves this by combining surface electromyography (sEMG) and accelerometer (ACC) signals using a feature fusion approach. The study introduces a multi-level attention mechanism for classification. This mechanism is based on convolutional neural networks (CNNs). During preprocessing, a local feature attention mechanism enhances local waveform features using the amplitude envelope. The model employs a dual-scale attention mechanism, which operates at both channel and neuron levels. These multi-level channel and neuron attention mechanisms improve the model's ability to learn from high-dimensional fused data, enhancing feature extraction and generalization. The local feature attention mechanism effectively boosts the model's classification accuracy and convergence, as evidenced by ablation experiments. The model, optimized with multi-level attention mechanisms, excels in accuracy and generalization, particularly in handling data with pseudo-artifacts. Computational analysis shows that the suggested optimization algorithm minimally affects CNN's training and testing times. In the end, the method in this study reaches recognition accuracies of 92.52%, 92.38%, and 92.30% and F1-scores of 91.92%, 92.13%, and 92.29% for the three fatigue states, confirming its reliability. This study offers technical support for the creation of affordable and reliable wearable motion monitoring devices.