چكيده به لاتين
One of the important applications of BCI is controlling of neuroprostheses to restore grasp to patients with paralyzed or amputated upper limbs. For reliable function of the neuroprosthese to get objects, pressing a button and generally doing precise movements, it is necessary for the user to control the amount of force necessary for grasping. For this reason, increasing the accuracy of continuous force decoding is an important issue for convenient function of these BCI systems. In most studies in the field of force decoding, linear models such as wiener filter, Kalman filter, PLS, etc. are used to decode force. So far, the effect of using nonlinear models that considers more complex relations between brain signal features is not investigated on force decoding. Our purpose in this thesis is to use several nonlinear models to decode force, evaluate and compare these methods with each other and provide a nonlinear model that will improve the accuracy of force decoding. For this purpose, we use kernel methods including kernel ridge, kernel PCR and kernel SVR, perceptron, RBF and ELM neural networks, generalized linear models and nonlinear PLS algorithms to decode force and compare their decoding accuracy. Among these methods, kernel ridge has the best performance.We discuss advantages and disadvantages of each of these methods and then, we propose a new method, nonlinear sparse PLS (NLS PLS) and compare its performance with previous methods. Our result showes that accuracy of force decoding using the NLS PLS method is more than other methods, and the mean correlation coefficient between the estimated and measured force is 0.73 and R2 is 0.63.
