Abstarct: The Brain Computer Interface (BCI) is a powerful system for communication between the brain and the outside world. Traditional BCI systems only worked on EEG. These signals, because of their benefits, facilitate the analysis and recording of mental activity. Recently, researchers have combined EEG signals to improve the performance of these systems with other signals. Among these signals, the combination of EEG and fNIRS signals has achieved favorable results. In most studies, only EEG or fNIRS signals are used as sequences of chain-like signals, and the complex correlation between adjacent signals, the time and position of the electrodes, is not considered. On the other hand, there are challenges that need to be addressed. The proposed method of this thesis is focused on two parts of its preprocessing for data preparation and classification. The preprocessing combines the spatial relationship between the EEG and fNIRS signals. This is done by using the position of the electrodes on the scalp. As a result, sequences of these chain-like signals can be converted into three-dimensional tensors. In this thesis, a deep neural network baxsed on two neural networks CNN and RNN has been proposed for accurate diagnosis of human mental goals using time and spatial features. The results of the proposed preprocessing method showed that the raw signals become images with more information. These images have spatial information of the signals at any moment in time. It was shown that the proposed model has a precision of 99.6% and outperform a set of baxseline methods and most recent deep learning-baxsed EEG or EEG-fNIRS recognition models, yielding a significant accuracy increase of 1 percent in the cross-subject validation scenario. On the other hand, the results showed that the proposed preprocessing and the conversion of raw signals to images with spatial information had a favorable effect on the deep neural network.