Improving Gesture Recognition for Amputees Based on Fusion of sEMG and Acceleration Signals Using Broad Learning System

Purpose Gesture recognition based on biological signals played a crucial role in the field of human-computer interaction. Despite achieving decent results in improving gesture recognition accuracy using different algorithms, significant challenges remained in enhancing recognition accuracy for multimodal signals and special patients such as upper limb amputees. Methods This paper proposed a Broad Learning System (BLS) method to recognize hand movements. The input signals fused surface electromyographic (sEMG) and acceleration signals, and the fused data were mapped into contour maps to extract features, which were transformed into two-dimensional grayscale images to achieve efficient encoding. The paper utilized the public databases "Ninapro DB2 and DB7" to evaluate the performance of the proposed method. Results Experimental results showed that the average gesture recognition accuracy of BLS reached 97.73%, with an average testing time of 0.093s, outperforming the other two algorithms, K-Nearest Neighbor (0.497s) and Binary Tree (2.412s). Moreover, the fused signals exhibited higher recognition accuracy for amputees. Conclusion In conclusion, the accuracy and real-time of the method could satisfy the requirements for controlling the prosthetic. Hence, it provided a reliable approach for research in fields such as rehabilitative medicine and prosthetic control.