A Novel AI-Based Detection Framework for Reused Cooking Oils: Advancing Food Safety and Cancer Prevention in Restaurant Foods

Graphical Abstract Abstract Cooking oil use increases the risk of cancer by producing carcinogenic polycyclic aromatic hydrocarbons (PAHs) and aldehydes, especially in developing countries. The detection techniques used today are time-consuming, reliant on laboratories, and incompatible with real-time monitoring. Here, we demonstrate how real-time cooking oil reuse detection is made possible with remarkable accuracy using hybrid deep learning architectures that combine ResNet50/VGG16 with Extreme Learning Machines with Local Receptive Fields (ELM-LRF). We examined 5,000 food samples made with oils that had been reused 0–3 times using multi-modal sensor fusion that included chemical gas sensors, visual imaging, and near-infrared spectroscopy. The ResNet50+ELM-LRF model successfully identified hazardous compound concentrations (PAHs >50 μg/kg, aldehydes >100 μg/kg) with a processing time of 2.3 seconds per sample, while the VGG16+ELM-LRF model showed 95.4% accuracy. An estimated 2.4 million diet-related cancer cases could be avoided each year in developing nations thanks to this portable CNN+ELM-LRF approach, which offers a revolutionary digital health solution for oil reuse detection in resource-constrained environments.