Lumbar EMG-Based Motion Intent Recognition for Industrial Exoskeletons

Motion intent recognition is critical for improving industrial exoskeleton performance and assisting individuals with disabilities. However, lumbar muscle EMG signals present challenges due to interindividual variability and limited motion range. To address this, we propose a two-step clustering method combined with FFT-based filtering to preprocess EMG data. A regularized LSTM-CNN model is then introduced, leveraging CNN for spatial feature extraction and LSTM for temporal dynamics modeling. Experimental results demonstrate 93% accuracy in motion intent recognition, confirming the model's reliability. This study provides an effective approach for enhancing exoskeleton control and assistive technologies.