چكيده به لاتين
Brain-computer interfaces are an auxiliary device for converting recorded brain signals into control signals of an artificial limb that will enable people with disabilities to regain power or control over the missing part. In order to have a good robotic arm, in addition to having complete motion information, the decoding of the force signal should also be made to determine the direction of the motive force. However, the higher the performance of this decryption, the better output will be seen. Spike based brain-computer systems usually use simple spike sorting methods. Given that the spikes are lost in the background during noise and the quality of the signal is reduced, providing an algorithm that improves the performance of the system in these situations will help to decode better. In this project, we intend to decode the force signal of the rats by detecting and sorting the recorded neuron spikes in the rat brain. For this purpose, an algorithm was first proposed for the spike sorting. In order to achieve the desired result, the proposed algorithm was tried to withstand signal changes due to the passage of time. With the help of it and after obtaining the firing rate, we decoded this signal. In the proposed method, optimal parameters are obtained for the best use of the wavelet algorithm, and with their help, a more favorable result is obtained in the clustering. These parameters are the parameters of scale and time transmission. The choice of optimal parameters is based on the fact that the maximum distance between the clusters in the channel is achieved. The proposed algorithm was first applied to simulated data and its performance was evaluated in the clustering of nerve spikes. After ensuring that its performance is correct, it was used to decode the actual data force signal. The firinf rate obtained from the proposed algorithm was decoded using two common methods of decoding partial least squares and Kalman filters in two different modes - the k-Fold and Leave-one Out methods. The results of our work showed that the applied force parameter can be decoded using the data recorded from the rat's brain region. In the analyzed data, partial least squares decoding algorithms showed better performance than the Kalman filter algorithm. Our study created a 4% increase in the correlation coefficient obtained using the proposed spike sorting algorithm compared to the multi-unit method. Based on our study, using the proposed algorithm and using the 7-Fold evaluation method, a correlation of 0.61 can be obtained. In the Leave-One Out evaluation method, a good correlation was also obtained, with the PLS decoding algorithm being 0.65 and Kalman filter algorithm, this number is 0.62 in the total of the three rats.
