Improving fine-grained food classification using deep residual learning and selective state space models.

Area of Science:

• Computational Nutrition
• Computer Vision
• Machine Learning

Background:

• Food classification is crucial for computational nutrition and food vision tasks.
• Fine-grained food classification presents challenges for traditional Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) due to complexity and computational cost.

Purpose of the Study:

• To develop an efficient and accurate food classification model.
• To address the limitations of existing models in handling complex food datasets.

Main Methods:

• Proposed the ResVMamba model, integrating a residual learning strategy within a state-space framework.
• Leveraged VMamba for efficient capture of global and local dependencies.
• Introduced and utilized the CNFOOD-241 food dataset for validation.

Main Results:

• ResVMamba achieved a Top-1 accuracy of 81.70% and a Top-5 accuracy of 96.83% on the CNFOOD-241 dataset.
• The model surpassed current state-of-the-art (SOTA) models in food recognition.
• Established a new performance benchmark for the CNFOOD-241 dataset.

Conclusions:

• Pioneered the integration of residual learning within VMamba for enhanced feature extraction.
• The ResVMamba model offers a computationally efficient and highly accurate solution for food classification.
• The study introduces the CNFOOD-241 dataset and provides open-source code for the ResVMamba model.