This paper presents a breast cancer image classification method that combines deep learning and transfer learning to address the challenges of limited samples, time-consuming feature design, and low accuracy in detecting and classifying breast cancer pathological images. The proposed algorithm is based on the DenseNet structure, which is enhanced with attention mechanisms to improve performance. The model is trained using multi-level transfer learning, leveraging pre-trained models from the ImageNet dataset and the LC2500 lung cancer dataset. The BreakHis dataset, containing 7,909 breast cancer images, is used for training, validation, and testing. Preprocessing techniques such as color normalization and data augmentation are applied to enhance the robustness and efficiency of the model. Experimental results show that the proposed method achieves a classification accuracy of over 84.0% on the test set, significantly outperforming previous models. The study highlights the effectiveness of deep transfer learning in improving the accuracy and efficiency of breast cancer detection, making it a valuable tool for medical applications. However, the model's limitations, such as its focus on binary classification and the need for further optimization, are also discussed.This paper presents a breast cancer image classification method that combines deep learning and transfer learning to address the challenges of limited samples, time-consuming feature design, and low accuracy in detecting and classifying breast cancer pathological images. The proposed algorithm is based on the DenseNet structure, which is enhanced with attention mechanisms to improve performance. The model is trained using multi-level transfer learning, leveraging pre-trained models from the ImageNet dataset and the LC2500 lung cancer dataset. The BreakHis dataset, containing 7,909 breast cancer images, is used for training, validation, and testing. Preprocessing techniques such as color normalization and data augmentation are applied to enhance the robustness and efficiency of the model. Experimental results show that the proposed method achieves a classification accuracy of over 84.0% on the test set, significantly outperforming previous models. The study highlights the effectiveness of deep transfer learning in improving the accuracy and efficiency of breast cancer detection, making it a valuable tool for medical applications. However, the model's limitations, such as its focus on binary classification and the need for further optimization, are also discussed.