Motion intention recognition of the affected hand based on the sEMG and improved DenseNet network

• DOI: 10.1016/j.heliyon.2024.e26763
• Publication Date: 2024-02-21T17:19:01Z
• Authors: Qunfeng Niu; Lei Shi; Yang Niu; Kunming Jia; Guangxiao Fan; Ranran Gui; Li Wang
• Keywords: Feature (linguistics)

The key to sEMG (surface electromyography)-based control of robotic hands is the utilization signals from affected hand amputees infer their motion intentions. With advancements in deep learning, researchers have successfully developed viable solutions for CNN (Convolutional Neural Network)-based gesture recognition. However, most studies primarily concentrated on utilizing data healthy subjects, often relying high-dimensional feature vectors obtained a substantial number electrodes. This approach has yielded high-performing recognition systems but failed consider considerable inconvenience that abundance electrodes poses daily lives and work patients. In this paper, we focused transradial used Ninapro DB3 database as our dataset. Firstly, introduce STFT (Short-Time Fourier Transform)-based time-frequency fusion map sEMG. includes both features localization signals. Secondly, propose an Improved DenseNet (Dense Convolutional Network) model recognizing intentions based Finally, addressing issue optimizing carried by amputees, PCMIRR (Pearson Correlation Motion Intention Recognition Rate) algorithm. algorithm optimizes channels considering Pearson correlation between rate single-channel data. experimental results reveal accuracy, recall, F1 score achieved were 93.82%, 93.61%, 93.65%, respectively. When was optimized 8, accuracy reached 94.50%. summary, paper ultimately attained precise amputees' while minimum channels. method offers novel sEMG-based bionic hands.