Abstarct: Brain-computer interface (BCI) systems are one of the topics that have attracted the attention of researchers and are used to establish a connection between brain signals such as EEG signals with a peripheral device such as an artificial hand or a wheelchair. BCI baxsed on motor imagery (MI-BCI) is one of the control methods in BCI research. In the MI-BCI method, a person imagines that he is moving one of his body parts, and this idea creates patterns in the brain signal. With the help of extracting these patterns and using pattern recognition methods, the movement imagined by the person can be recognized. The aim of this thesis is to classify the EEG signals related to the movement perception of the right hand, left hand, legs and language of BCI Competition IV 2a dataset. In this work, features extracted from the frequency and time domains using the filterbank method are concatenated with features extracted from the statistical or spatial domain using the CSP method. After concatenating, these features have been classified by the CNN networks. In this research, several different structures have been implemented to classify this dataset by CNN networks. The results are compared with the results of the best-studied work closely related to this work performed with the same dataset. The results show that the concatenation of spatial features with raw data in CNN networks does not improve the classification result; but using the fusion of the classification results of raw data in a CNN network and spatial data with the help of KNN classification, the result of data classification can be improved by 76.35%. For fusion of the results of two classifications, the confidency criterion obtained for each of the classifications has been used in the classification of each data.