With the rapid development of medical imaging analysis, automatic classification methods for pathological images have become important tools for advancing clinical pathology practice. However, the lack of pixel-level annotations has become a key bottleneck limiting the development of these methods. To address this issue, this project proposes a weakly supervised multi-instance learning method focusing on pathological images of lung and breast cancer. By leveraging long-sequence modeling based on a state-space dual model and multi-sequence feature fusion, the method achieves accurate classification performance. The proposed multi-sequence fusion method effectively utilizes both sequence-dependent and sequence-independent features, providing more comprehensive information for pathological image classification. The results show that the model performs excellently in pathological image classification tasks, achieving an AUC value of 95.3% and a classification accuracy of 87.9% in the classification of lung adenocarcinoma and squamous cell carcinoma. This achievement provides strong support for the automatic classification of pathological images and holds significant clinical application value.