In this study, we used a high-sensitivity multichannel fNIRS system based on the lock-in photon-counting technique to detect signals in the human brain by employing the full parallel excitation method. We used the original light intensity and hemoglobin concentration data to construct the SVM model to recognize subjects’ "positive/negative" binary intention. The average classification accuracies of the original light intensity data and hemoglobin concentration change data were 70.6%±3.7% and 73.1%±1.7%, respectively. Therefore, we demonstrated the ability of the high-sensitivity multichannel fNIRS system to directly detect the "positive/negative" binary intention of the human brain. The findings of this study provide a useful idea for applying fNIRS-BCI in clinical applications and daily lives, such as helping patients with locked-in syndrome express their intention more directly and developing a more convenient brain-controlled smart home.