Lower Back Muscle Fatigue Recognition Based on the Fusion-Information of Multi-Channel sEMG and NIRS Simultaneous Recordings.

Accurate recognition of muscle fatigue in the lower back is essential for preventing low back pain and reducing the risk of occupational injuries. However, current recognition accuracy remains unsatisfactory due to limitations in both measurement tools and recognition algorithms. This study proposes a novel recognition framework for muscle fatigue based on a 60-channel hybrid physiological sensor array and a deep learning algorithm. The sensor array, which integrated surface electromyography (sEMG) electrodes and near-infrared spectroscopy (NIRS) probes, enabled simultaneous, co-located recording of multimodal topographic signals that reflect both neuromuscular and hemodynamic activity. To effectively fuse and analyze this multimodal information, a dual-stream convolutional hybrid attention network (DCHANet) was developed and evaluated under both subject-wise and cross-subject conditions. The network comprises two modality-specific feature-extraction streams tailored to the characteristics of sEMG and NIRS, which were subsequently fused by a hierarchical attention-fusion module. Recognition performance was assessed on three-level (FAT-3) and fifteen-level (FAT-15) fatigue recognition tasks. With multimodal (sEMG-NIRS) input, DCHANet achieved high classification accuracy in the FAT-3 task (subject-wise: 97.93%, cross-subject: 96.80%) and the FAT-15 task (subject-wise: 91.06%, cross-subject: 88.53%). Compared with unimodal inputs, including sEMG and NIRS alone, the multimodal DCHANet achieved higher accuracy. It also outperformed conventional machine learning methods based on histogram of oriented gradients and standard convolutional neural network. These findings highlight the potential of combining hybrid physiological sensing with attention-based deep learning for precise and fine-grained recognition of lower back muscle fatigue, offering a promising solution for clinical monitoring and early intervention in musculoskeletal disorders.