A Novel Approach to Window Parameter Selection in sEMG-Based Intention Recognition

Surface electromyography (sEMG) signals are generated before the actual movement of the human body begins, playing a significant role in predicting human motion for exoskeleton control. A sliding window with parameters such as the length of the window and the size of the overlap is commonly used to process sEMG signals. Most studies suggest that longer window length provide more accurate statistical features but result in fewer samples, while larger overlap size improve performance but lead to more similar sample characteristics. Existing methods for selecting window length and overlap size typically require processing window data features and training multiple classifiers on different datasets to find the optimal combination. However, these methods are time-consuming and lack interpretability in practical applications. This study investigates the relationship between window length, overlap size, and classification results. It also introduces a sample distribution window metric algorithm to select the optimal window length and overlap size for sEMG-based Intention recognition. After sampling each parameter combination with a minimal number of samples, we use the trace of the between-class matrix and within-class matrix as the criterion to identify the optimal parameter combination. Our approach does not require training feature data to find the optimal parameters, and experimental results on publicly available datasets demonstrate the effectiveness and practicality of this algorithm.