As a pivotal area of research in the field of computer vision, the technology for food identification has become indispensable across diverse domains including dietary nutrition monitoring, intelligent service provision in restaurants, and ensuring quality control within the food industry. However, recognizing food images falls within the domain of Fine-Grained Visual Classification (FGVC), which presents challenges such as inter-class similarity, intra-class variability, and the complexity of capturing intricate local features. Researchers have primarily focused on deep information in deep convolutional neural networks for fine-grained visual classification, often neglecting shallow and detailed information. Taking these factors into account, we propose a Multi-level Attention Feature Fusion Network (MAF-Net). Specifically, we use feature maps generated by the Convolutional Neural Networks (CNNs) backbone network at different stages as inputs. We apply a self-attention mechanism to identify local features on these feature maps and then stack them together. The feature vectors obtained through the attention mechanism are then integrated with the original input to enhance data augmentation. Simultaneously, to capture as many local features as possible, we encourage multi-scale features to concentrate on distinct local regions at each stage by maximizing the Kullback-Leibler Divergence (KL-divergence) between the different stages. Additionally, we present a novel approach called subclass center loss (SCloss) to implement label smoothing, minimize intra-class feature distribution differences, and enhance the model’s generalization capability. Experiments conducted on three food image datasets—CETH Food-101, Vireo Food-172, and UEC Food-100—demonstrated the superiority of the proposed model. The model achieved Top-1 accuracies of 90.22%, 89.86%, and 90.61% on CETH Food-101, Vireo Food-172, and UEC Food-100, respectively. Notably, our method not only outperformed other methods in terms of the Top-5 accuracy of Vireo Food-172 but also achieved the highest performance in the Top-1 accuracies of UEC Food-100.

Food image recognition is a key research area in food computing, with applications in dietary assessment, menu analysis, and nutrition monitoring. However, imaging devices and environmental factors introduce noise, limiting classification performance. To address this, we propose a food image recognition method based on anti-noise learning and covariance feature enhancement. Specifically, we design a Noise Adaptive Recognition Module (NARM), which incorporates noisy images during training and treats denoising as an auxiliary task to enhance noise invariance and recognition accuracy. To mitigate the adverse effects of noise and strengthen the representation of small eigenvalues, we introduce Eigenvalue-Enhanced Global Covariance Pooling (EGCP) into NARM. Furthermore, we develop a Weighted Multi-Granularity Fusion (WMF) method to improve feature extraction. Combined with the Progressive Temperature-Aware Feature Distillation (PTAFD) strategy, our approach optimizes model efficiency without adding overhead to the backbone. Experimental results demonstrate that our model achieves state-of-the-art performance on the ETH Food-101 and Vireo Food-172 datasets. Specifically, it reaches a Top-1 accuracy of 92.57% on ETH Food-101, outperforming existing methods, and it also delivers strong results in Top-5 on ETH Food-101 and both Top-1 and Top-5 on Vireo Food-172. These findings confirmed the effectiveness and robustness of the proposed approach in real-world food image recognition.

Food image recognition is a key research area in food computing, with widespread applications in dietary assessment, menu analysis, and nutrition monitoring. However, due to the influence of imaging devices and environmental factors, this technology is often subject to varying degrees of noise interference during application, which limits classification performance. To address this issue, we proposed a food image recognition method based on anti-noise learning and covariance feature enhancement. We designed a Noise Adaptive Recognition Module (NARM), which inputs additional noisy images during model training, treating denoising as an auxiliary task to progressively enhance the model's noise invariance and improve recognition accuracy. To mitigate the potential negative effects caused by the additional noise and to enhance the representation power of small eigenvalues, we innovatively proposed the Eigenvalue-Enhanced Global Covariance Pooling (EGCP) and integrated it into NARM. Simultaneously, we designed a new Weighted Multi-Granularity Fusion (WMF) method to improve feature extraction capability progressively. Additionally, by combining the Progressive Temperature-Aware Feature Distillation (PTAFD) method, we further optimized model efficiency without adding extra overhead to the original backbone network. The experimental outcomes show that the model we have developed delivers top-tier performance on the ETH Food-101 and Vireo Food-172 food image datasets. Specifically, our method achieves a Top-1 accuracy of 92.57 on ETH Food-101, surpassing other existing methods, and also performs well on the Top-5 accuracy on ETH Food-101 and the Top-1 and Top-5 accuracy on Vireo Food-172.