Enhancing Hospital Meal Safety: Integrating Adaptive Priority Dual Loss and False Negative Weighting in CNNs for Allergenic Food Recognition

Ensuring meal safety in hospital settings is of paramount importance, given the specialized dietary needs and the vulnerability of patients. This paper addresses the critical need for accurate food component identification in hospital food service systems to prevent adverse reactions due to allergens or dietary non-compliance. We propose a novel approach to enhance automated meal recognition systems, focusing on the recognition of allergenic food components with priority. We integrate two novel methods within Convolutional Neural Networks: Adaptive Priority Dual Loss Function Strategy and False Negative Weighting with Hyper-parameter Calibration. The results show an improvement in the average recall rate for priority food categories, thereby enhancing meal safety. We compared our method with traditional techniques using benchmark neural network architectures. The findings reveal that our approach significantly improves the recognition rates of priority classes, which is crucial in a hospital setting where accurate food identification can have profound health implications. This research contributes to the field by ofering an innovative solution to a pressing food safety challenge, paving the way for safer and more efficient food service systems in hospitals.