EMG Based Rehabilitation Gesture Recognition Using DAE-CNN-LSTM Hybrid Model

Stroke, as one of the leading causes of long-term disability globally, often results in motor impairments, particularly in the hands, significantly affecting patients' daily activities and causing profound psychological trauma. Rehabilitation gesture recognition, as one of the key means of acti...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2024 World Rehabilitation Robot Convention (WRRC) s. 1 - 6
Hlavní autoři: Cao, Wujing, Guo, Xinqiang, Zou, Yupeng, Zhang, Shuo, Luo, Mingxiang, Kobsiriphat, Worawarit, Wu, Xinyu, Yin, Meng
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 23.08.2024
Témata:
Accuracy
CNN-LSTM algorithm
Convolutional neural networks
Deep learning
Electromyography
Feature extraction
Gesture recognition
Long short term memory
Noise
Rehabilitation
Stroke
Stroke (medical condition)
Support vector machines
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Stroke, as one of the leading causes of long-term disability globally, often results in motor impairments, particularly in the hands, significantly affecting patients' daily activities and causing profound psychological trauma. Rehabilitation gesture recognition, as one of the key means of active rehabilitation medicine, holds significant potential in stroke rehabilitation, providing real-time feedback on patient progress and enabling personalized interventions to meet individual needs. Traditional gesture recognition methods face numerous challenges when dealing with stroke patient data, such as noise, motion blur, and individual differences. To address these challenges, this paper proposes a novel approach for stroke patient gesture recognition using deep learning models. We adopt a strategy combining denoising autoencoder (DAE), convolutional neural network (CNN), and long short-term memory (LSTM) to enhance the accurate recognition of hand movements in stroke patients. Specifically, DAE is utilized for denoising EMG signals, extracting features, and reducing noise to improve the signal's noise resistance. CNN is employed for spatial feature extraction, while LSTM captures temporal dependencies in gesture sequences. By integrating these three deep learning models, our aim is to enhance the accuracy and robustness of rehabilitation gesture recognition. We validate the proposed method's effectiveness on an EMG dataset from seven subjects through experiments and compare it with traditional machine learning and individual CNN and LSTM algorithms. Experimental results demonstrate significant performance improvements of our hybrid model in rehabilitation gesture recognition tasks, indicating promising application prospects and practical significance.
DOI:10.1109/WRRC62201.2024.10696763