The invention relates to a brain function connection network analysis method based on a resting state functional magnetic resonance image, and the method is characterized in that the method comprises the following steps: S1, obtaining an rs-fMRI image, and carrying out the preprocessing of the rs-fMRI image; s2, segmenting the rs-fMRI image by adopting three brain templates with different scales, and constructing three graphs G1, G2 and G3; s3, according to the obtained three graphs G1, G2 and G3, hidden feature representation of the graphs is learned by adopting a GNN module, and preliminary prediction labels L1, L2 and L3 are obtained; obtaining a final prediction label after voting; s4, performing significance analysis on a pooling result of the GNN module in the step S3, and obtaining a brain region with significant difference in the functional network; and S5, mapping the prominent brain region of the functional network to the Yeo 7 brain functional network map, and obtaining the mapped functional network connection, namely the individual difference functional sub-network. According to the method, the brain network is constructed by using the three different scales of brain templates, the integrated learning strategy is adopted, the multi-scale information of the multiple brain templates is fully fused, and the model performance is improved.