This study reveals the adaptive phenomenon in the fNIRS brain activation experiment and the factors that influence it. It is suggested that the research of BCI based on fNIRS also needs to consider adaptability. Since the activation level is not stable at the initial stage of the experiment, the magnitude of the amplitude change may influence the final feature extraction and classification effect. Therefore, it is necessary to conduct a period of pre-experiment and process the experimental data before the start of the experiment and wait for the amplitude to stabilize. If this characteristic is not considered, it will be difficult to extract the stable signal characteristics of fNIRS, which will greatly reduce the recognition accuracy of BCI. Although there are individual differences in this adaptability and it is not monotonously decreasing, this may involve more complex neural mechanisms. However, most people reach a relatively stable level after 3 d. This result suggests that we can consider using the data after 3 days of training when conducting the fNIRS-based brain-computer interface research. At the same time, the difference in training time requires that in the experimental process, the number of experimental days and daily experiment volume of each subject should have a certain consistency so as to eliminate the errors caused by different degrees of adaptation. Different paradigms have different adaptation time. Therefore, unless the effects of different difficulty paradigms are studied, it is necessary to ensure that the experimental paradigms are similar in difficulty to improve the efficiency of adaptive training and eliminate the errors of different paradigms.