Brain-computer interface technology (BCI), as the key technology to connect human brain with external interface technology, has received extensive attention from more and more scholars in recent years. Research on motion detection and behavior control using BCI has become the focus of BCI system. However, with the increasing complexity of motion, higher requirements are put forward for BCI system. The current research on electroencephalogram feature extraction and classification under complex motion tasks still has the problems of simple experimental paradigm and low complexity in motion-related tasks. Based on this, this paper conducts the electroencephalogram acquisition experiment under complex motion tasks, and the installation and debugging of experimental equipment as well as the experimental design are completely in accordance with international standards. At the same time, the functional networks constructed by Pearson coefficient matrix and adjacency matrix were used to extract the features of EEG, and the extracted features were classified into electroencephalogram signals under complex motor tasks by the constructed Graph Convolutional Neural Network (GCNs-Net) model, which comprehensively evaluated the classification accuracy by electroencephalogram and behavioral indicators, and provided a feasible idea and method for the field of stability and accuracy of electroencephalogram identification in brain computer control.