چكيده به لاتين
Today, human-computer communication systems are very important in most scientific fields, especially the control of prostheses. Therefore, this research examines the issue of continuous decoding of hand movement parameters in three main dimensions with the help of signals received from the brain using the ECOG method. The received signals are related to brain activities and are the main tools for detecting the type of goals in the user's mind, which are collected with the help of electrodes implanted in the epidural part of the brain. There are many methods for detecting and decoding brain signals, and due to the increasing progress of artificial intelligence in the world, we have used deep learning methods in this research. In the presented methods, inspired by classical methods and famous neural networks, two deep neural networks have been designed, the first network consists of a convolutional part and a recurrent part for decoding continuous hand movement, and the second network is a deep network with more depth, which It is designed based on the deep network of Rosenet. In the proposed methods, feature extraction is performed by the convolutional part, and the recursive part will decode the hand movement parameters based on the calculated features. To evaluate the model, the data set received from two monkeys that are changing the direction of their hand movement in three different dimensions have been used. evaluation criteria of correlation coefficient, normalized mean squared error, and squared mean squared error have been used to evaluate the proposed models. The results obtained from the proposed methods have been compared with the methods available in the references. The proposed methods are new compared to the classical methods used on these data, and after receiving the raw data, all feature extraction and decoding steps are performed automatically. The proposed methods of this research have provided acceptable accuracies compared to previous studies.
