A Brain Tumor Image Classification Method Based on Improved MobileViT Model

To address the issues of redundant parameters and low training efficiency in deep learning models for image recognition tasks, this paper proposes a lightweight image classification model based on an improved MobileViT architecture. First, the Efficient Channel Attention (ECA) mechanism is integrated into the MobileViT backbone to enhance the network's ability to capture critical channel-wise features and improve feature representation. Second, for the optimizer, the original AdamW is replaced with Stochastic Gradient Descent (SGD) combined with a cosine annealing learning rate scheduling strategy. This modification not only enhances the model's training stability but also accelerates convergence. Finally, a comparative experiment is conducted using a real brain tumor dataset. The results demonstrate that, compared with the original MobileViT and other mainstream models, the proposed approach achieves higher recognition accuracy while maintaining low computational complexity. These advantages make it highly suitable for practical applications, particularly in edge computing scenarios with limited computational resources.