Food image recognition plays a crucial role in food safety monitoring, nutritional analysis, and dietary recommendation systems. However, the diversity, complexity, and external factors such as lighting conditions pose numerous difficulties and challenges to the recognition task. In order to address these issues, this paper proposed a food image classification algorithm based on improved MobileNetV3-Large. Firstly, building upon the pre-trained MobileNetV3-Large model, the PReLu activation function and NAM attention mechanism were introduced to enhance the model's focus on key features by capturing non-local dependencies in images. Subsequently, a multi-task loss function was incorporated to further improve the classification performance by simultaneously optimizing multiple related tasks. Finally, the TrivialAugment data augmentation technique was employed to enhance the model's generalization ability by expanding the scale and diversity of the training dataset. Experimental results demonstrated that through these improvements, the model's accuracy on the Food-101 dataset increased from 66.9% to 84.2%, demonstrating the effectiveness of the proposed approach.