چكيده به لاتين
Model-based myoelectric control has been proposed as a valuable technique for controlling artificial hands in people with amputation over the past two decades. Despite the variety and breadth of algorithms currently proposed for controlling myoelectric patterns based on pattern recognition, barriers and challenges remain as unresolved issues for this control plan for commercial and clinical implementation. One of the most important challenges in myoelectric control is the sensitivity of the algorithm to the position of the hand. The approaches that previous studies and articles have suggested for this challenge can be categorized into two categories: 1. Using the accelerometer sensor to determine the position of the hand and 2. Training users in multiple positions. The first approach is to use an external sensor and can not be at least cost-effective, and the second approach requires training and calibration time for user training. So, we study two main objectives in this project, the first goal is to determine the position of the hand without using any external sensors and only with its own electromyograph data (EMG), and the second objective Classification of moving classes in a given position with the minimum amount of training and calibration data is proposed. Therefore, in this study, we propose a two-dimensional algorithm that firstly identifies the position of the hand, and secondly, to reduce the learning and calibration time using an adaptive algorithm using the data of other users for learning. This comparative algorithm obtains a linear transformation for transferring a feature space from one user to another by minimizing a statistical distance between classes in a feature space. In the results section, the position of each of the positions was identified in 4 different positions with an average accuracy of 87.21%, and the determination of the pattern recognition algorithm in 8 different movements of the wrists by using patterns matching learning from other users with a precision of 83.21%. The results show that we have been able to identify the position of the hand without using external sensors and using our own EMG data and use the hand position information and other users data with the proper precision to classify the eight classes of motion.
Keywords: Myoelectric control, Electromyograph, Accelerometer, Adaptive algorithm and Training time.
