Subspace and second-order statistical distribution alignment for cross-domain recognition of human hand motions

Collaborative robots are an integral component of the intelligent manufacturing field. The recognition of hand motions based on surface electromyography signals is even more significant for advancing the research of collaborative robots. However, supervised learning based hand motions recognition methods require an enormous quantity of data as support and must meet the condition of independent and identical distribution. In real-world scenarios, gender classifications, collecting conditions, and even individual-to-individual variances may influence the data, posing a challenge for the recognition of hand motions from new participants. Consequently, we propose a domain adaptive framework based on subspace and second-order statistical distribution alignment (SSDA) to overcome the issue of non-independently and identically distributed data shift. We combine the second-order statistical distribution alignment and subspace alignment. SSDA diminishes the geometric and statistical distribution discrepancies between the training and test sets in hand motions recognition. SSDA enhances the average accuracy of hand motions recognition by 26.66%, 14.43%, and 25.76% in three cross-domain scenarios (cross-gender, cross-circumstance, and cross-individual), respectively, compared with the baseline method of direct classification. Experimental results indicate that the proposed method is effective in solving the problem of distribution shift between target data (test set) and priori data (training set) in hand motions recognition. Simultaneously, it improves the recognition accuracy of classifier for different distributed data, thereby providing a new idea for achieving efficient human-robot collaboration.