Rapid advances in deep learning enabled us to develop various brain-computer interface (BCI) applications. This study presents a novel BCI framework in virtual-reality environment based on an electroencephalography (EEG) decoder for learning strategy classification. We collected 9 subjects' EEG data using a 6-channel EEG-VR headset, and implemented 2D convolutional neural networks with spectrograms as input features. The proposed method achieved 82% classification accuracy, despite a small number of channels and noise artifact. The results suggest a possibility of exploiting BCI technologies for various VR applications.