Aiming at the problem of poor consistency between the enhanced samples and the original samples in the current data enhancement methods. In this paper, we propose a data enhancement method with improved symplectic geometry reconstruction. The method expands a sufficient number of augmented samples while ensuring high similarity to real fault samples. First, the components to be augmented and their weights are selected from the decomposed symplectic geometric modal components using a random sampling method. Second, a data-weighted scaling criterion is formulated to augment the modal component amplitudes. Then, the summation and reconstruction of all amplitude enhancement components are performed on the basis of ensuring that the statistical characteristics of the enhanced samples remain consistent with the original samples. Finally, the proposed method is combined with the deep residual network, and the experimental dataset of hydraulic pump failure simulation is utilized to verify the effectiveness of the proposed method in obtaining the enhanced fault sample information. The diagnostic results show that the fault diagnosis model established by the method has higher accuracy and convergence ability in hydraulic pump data imbalance fault diagnosis compared with other comparative models.