The domain generalization for digital pathology images is an important and challenging problem in computational pathology. Among them, the domain generalization method based on uncertainty modeling is an important way to solve it. However, current methods based on uncertain modeling have problems such as the inability to model the statistical characteristics of global feature distributions and the inability to align multi-source domain feature distributions. Therefore, we propose a domain-invariant feature-enhanced domain generalization method based on Collaborative Feature Matching and Uncertainty Perturbation. Our method first achieves alignment of multi-source domain feature distributions through multiple collaborative feature matching mechanisms to reduce distribution differences. Then, domain-specific features are weakened through adversarial learning while preserving domain-invariant features. Finally, perturb the global feature statistics to expand the hypothesis space for the distribution of the target domain. We conducted extensive experiments on representative domain generalization datasets such as Camelyon17, JMI, and IHC4BC. The experimental results indicate that we achieve state-of-the-art performance compared to previous methods. It proves the method can effectively alleviate the domain shift problem of multi-source domain digital pathological images and improve the model’s generalization.