In view of the challenges of high-order functional connectivity modeling and classification in functional magnetic resonance imaging (fMRI), this paper proposes a structural hierarchical mild cognitive impairment (MCI) recognition framework that integrates the dual-stage aligned synergistic connectome (DAS-Connectome) and the structural embedding refinement layer (SERL). In the process of high-order construction, the DAS-Connectome introduces a two-level structural modeling mechanism: First, the neural resonance matrix is standardized and preprocessed by normalized denoising operation to construct a stable high-order neural dependency structure; Then, the structural mapping enhancement is performed to couple the high-order structure with the original low-order matrix to generate the DAS-Connectome, thereby integrating the cross-order synergistic relationship and improving the consistency and discriminant density of the structural expression. Subsequently, the SERL is constructed to extract the low-dimensional representation of the DAS-Connectome with the help of the pre-trained Vision Transformer (ViT). Finally, the embedded features are input into a lightweight classifier to complete the MCI discrimination. Experiments show that the classification accuracy of the proposed method is improved by up to 16% compared with the traditional method on the ADNI dataset, and it shows good stability and generalization ability under small sample conditions.