Date of Award
Doctor of Philosophy (PhD)
Food safety and foodborne diseases are significant global public health concerns. Precise, reliable, and speedy contamination detection and disinfection technology while preserving the business owners’ data privacy is an ongoing challenge for the food-service industry. Contamination in food-related services can cause foodborne illness, endangering customers and jeopardizing provider reputations.This dissertation performed a cleanliness assessment and disinfection and data privacy assurance in the food services industry using fluorescence imaging, state-of-the-art deep learning algorithms, and a novel paradigm in machine learning named federated learning. In chapter 3, we combined two deep learning algorithms (EfficientNet-B0 and U-Net) and fluorescence imaging for automatic detection and precise segmentation of fecal contamination on meat carcasses to provide higher levels of food safety assurance in meat processing facilities. We achieved a 97.32% accuracy for discriminating between clean and contaminated areas on carcasses and an intersection over union (IoU) score of 89.34% for segmenting areas with fecal residue. In chapter 4, we focused on cleanliness assessment and disinfection of organic residue-based contamination in institutional kitchens and restaurants. We used new fluorescence imaging technology, applying Xception and DeepLabv3+ deep learning algorithms to identify and segment contaminated areas in images of equipment and surfaces. Deep learning models demonstrated a 98.78% accuracy for differentiation between clean and contaminated frames on various surfaces and resulted in an intersection over union (IoU) score of 95.13% for the segmentation of contamination. Further, in chapter 5, the main focus of the study was to address the concerns regarding using new technologies that can increase privacy risks and leaks of sensitive information. Hence, we used federated learning as a new paradigm in machine learning combined with fluorescence imaging technology and two deep learning models, including MobileNetv3 and DeepLabv3+, to identify and segment the contaminated area on different equipment and surfaces. The model was trained and validated on the data of eight clients and tested on two new clients' data. The model achieved a 95.83% and 94.94% accuracy (F-scores of 96.15% and 95.61%) for classification between clean and contamination frames of the two new clients and resulted in an intersection over union (IoU) score of 91.23% and 89.45% for segmentation of the contaminated areas. Overall, the findings demonstrate that fluorescence imaging combined with state-of-the-art deep learning models not only can improve safety and cleanliness assurance but also ensure client data privacy.
Taheri Gorji, Hamed, "A Deep Learning-Based Framework For Food Safety, Cleanliness, And Data Privacy Assurance" (2022). Theses and Dissertations. 4560.
Available for download on Thursday, January 18, 2024