<p>As an important research direction in the fields of computer vision and machine learning, image recognition faces unique challenges in food image recognition, especially for the recognition task of Chinese food. Due to the profound and extensive Chinese culinary culture and the wide variety of dishes and cooking methods, recognition of the image of Chinese food has become a research difficulty in this field. Specifically, many Chinese dishes have significant differences in nutritional value, but their appearance features are extremely similar, which poses a huge challenge to traditional recognition methods. In response to this issue, this article constructs a novel Chinese food image recognition method by deeply fusing convolutional neural networks capable of extracting deep local features with Swin Transformer. This method can effectively capture fine-grained features of food and generate more discriminative enhanced feature representations. In addition, this article also designs an end-to-end classifier that can efficiently filter out the most discriminative information from enhanced features, thereby improving recognition accuracy. This method performs well in dealing with foods with similar appearances but different nutritional characteristics and is expected to provide reliable technical support for applications such as smart diet recommendations and health monitoring. This article has been fully validated on the public Chinese food datasets ChineseFoodNet and CNFOOD-241. The results showed that the recognition accuracy of this method on these two datasets reached 85.70% and 84.58%, respectively, significantly better than existing supervised learning methods, demonstrating stronger feature learning ability and generalization performance. This achievement provides a new solution for the recognition of Chinese food images and lays an important foundation for subsequent research in related fields.</p>