سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Investigating a Real-time sEMG-based Approaches for Grasping Recognition

نویسندگان :

Monire Ameri Haftador¹ Ali Akbari² Mehran Jahed³

1- دانشگاه صنعتی شریف، تهران، ایران 2- دانشگاه صنعتی شریف، تهران، ایران 3- دانشگاه صنعتی شریف، تهران، ایران

کلمات کلیدی :

Hand Grasping Recognition،Surface Electromyography (sEMG)،Real-Time Systems،Short-time Fourier Transform (STFT)،Convolutional Neural Networks (CNN)

چکیده :

To fully exploit real-time prosthetics and exoskeleton assist devises, human-machine interfaces that can effectively deduce related activity and intent are essential. Surface electromyography (sEMG) provides a well-established non-invasive method for this purpose, yet two key barriers to its broad adoption are attaining recognition latency well below 200ms and preserving accuracy in presence of signal drift. In order to describe an orderly solution to these issues, this paper is a comprehensive assessment of deep learning and conventional algorithms. To establish a comprehensive scheme for targeted gesture, data was meticulously collected from individuals in the biomedical engineering lab. Furthermore, to be able to contrast the proposed method against the already reported work, a well-established publicly available dataset, namely EMG-EPN-612 was utilized. To achieve appropriate real-time accuracy, commonly used classifiers, namely Support Vector Machine (SVM), Random Forest, and Convolutional Neural Networks (CNN) were implemented and compared based on these metrics. Input was rigorously evaluated in three forms, processed signals, handcrafted features, and Short-Time Fourier Transform (STFT) images, in a bid to determine the optimal strategy. Although all these models were shown to support the required real-time constraint, however only the CNN model applied to the STFT inputs achieved the acceptable 92% accuracy on the EMG-EPN-612 dataset, as compared to SVM applied to handcrafted features of 84% accuracy on the recorded dataset. These results provide first-time explanation and trade-off between model complexity and computation cost, and required accuracy. This research provides useful recommendations that further assist in developing more effective, responsive, and accessible hand assist devices and prosthetics.