Automated Classification of Limb Movements Using Time–Frequency Features and Multi-Optimizer Ensemble Framework

The electromyography (EMG) signals used in the recognition of limb-movements are crucial to intelligent prosthetic and rehabilitation systems. The existing techniques have the drawbacks of signal instability, overlapping features and poor extrapolative capability between subjects and orientations. This paper suggests a time-frequency meta-heuristic optimizer based machine-learned pipeline to automatically classify limb motions. The framework first displays time frequency separation in order to recover dynamic contents of non-stationary EMG signals. It then uses a collection of 44 different meta-heuristic optimizers in order to identify stable and discriminative subsets of features. A strategy based on voting picks the features, which uniformly enhance performance over optimizers. Two public datasets, FORS -EMG (8 channels, 12 gestures, 3 orientations) and UCI sEMG Basic Hand Movements (2 channels, 6 gestures) are tested on the proposed method through cross-subject and cross-orientation validation. Through the results, it is found that the model has a better classification accuracy, greater feature stability, and lower computational complexity than the current single-optimizer methods. The size and weight of the framework are small enough to fit the framework in real-time applications of embedded prosthetic control and rehabilitation devices. This study introduces a general and decipherable course of action of EMG-based limb-movement recognition frameworks that can be customized to various individuals and recording scenarios.