In today's world, problems and neurological diseases that lead to immobility can be solved to a large extent with the help of brain and computer interface technology. This technology predicts people's movement decisions by analyzing brain signals and sends them to the robot. Ensemble Empirical Mode Decomposition (EEMD) is a method that transforms input data into Intrinsic Mode Functions (IMF). In addition, independent component analysis (ICA) and correlation component analysis (CCA) methods are among the most common and fastest methods of blind analysis of the sources of a signal. In this research, the sEEMD-ICA-CCA method, which is a three-step integrated method for data preprocessing, is used to preprocessing of brain data. In this method, first, the data is transformed from a data form with multiple channels to a vector according to the serialization algorithm, then this vector is transformed into a set of intrinsic mode functions with the EEMD algorithm, and then the intrinsic mode functions that have a lower correlation coefficient, go through the ICA algorithm and identify the blinde sources in it and removes them. In the next step, the vector is entered into the CCA algorithm, and using of this method, the intrinsic mode functions that passed through the previous step are classified based on the autocorrelation coefficient, and the category that has a lower correlation than the rest of the intrinsic mode functions are identified as artifacts and removed. These denoised eigenstate functions are reassembled and new clean data is created. The energy of these data is calculated and according to the energy changes in each data channel and each electrode, features are extracted that by comparing them with each other, different body movements are identified in each person. Three different categories of data are used to eva‎luate the accuracy of this method. The first category is synthetic data with 31 channels, the second category is the data available on the internet portal of BCI competitions with 5 audiences and the data of two categories, and the third category is 6 series of practical tests with different movements of hands and feet from 40 samples(15 women and 25 men). The performance of the proposed pre-processing algorithm in synthetic data and BCI competition data was eva‎luated by RRMSE criterion. With the increase of SNR in all types of data, RRMSE decreased and reached 0.3, which showed good performance to clean the data. Then, the data of BCI competitions and the experimental data obtained after pre-processing by the desired method, were placed in the classification and machine learning algorithm, and the classification was done, and the desired movement was performed during the label test, was established and estimated. These labels, which are the result of the estimation of the complete BCI algorithm, are introduced as a pre-defined control command for the robot and the robot performs the desired movement. This algorithm provided about 70% accuracy in the test data despite the presence of noise.

محمد دانش

عرفان نقش

سعيد بهبهاني , مهدي كشميري