Real-time decomposition method, system, terminal and storage medium for dynamic shrinking surface myoelectric signal based on target detection transformer

Abstract

This invention relates to the field of physiological signal decomposition technology, and discloses a method, system, terminal, and storage medium for real-time decomposition of dynamically contracting surface electromyography (EMG) signals based on a target detection transformer. The method models EMG signal decomposition as a multi-target detection task on a one-dimensional sequence, constructing an end-to-end decomposition network based on a target detection transformer, including a feature extraction module, an encoding / decoding module, and a prediction head. Parallel detection of asynchronous motor unit discharge events is achieved through learnable query vectors. During real-time decomposition, the EMG signals in a sliding window are input to multiple trained decomposition models, and the output discharge events are filtered and fused to obtain the final motor unit pulse sequence. This invention can adapt to non-stationary characteristics such as action potential waveform drift during dynamic contraction, solving the problems of decreased decomposition accuracy and insufficient real-time performance under dynamic contraction conditions, and achieving high-precision, low-latency real-time decomposition of surface EMG signals.

Images