The human brain functional connectivity detection is an important technique in neuroscience research. The restricted boltzmann machine (RBM), modeling on a large amount of multi-subject functional magnetic resonance imaging (fMRI) data, it can discover the brain functional connectivity. However, the former method with restriction of the huge training data, it can not detect the functional connectivity on single-subject data effectively. In this research, a novel functional connectivity detection model taking advantage of the sparsity is presented, which is an effective combination of the spatial-domain sparse approximation theory and the RBM technique. The experimental results demonstrated that the proposed model could effectively discover both the temporal dynamic model and the corresponding spatial functional maps on the single-subject data, which settled the the bottleneck of RBM.