Progress 07/01/23 to 03/29/24

Outputs
Target Audience:Food safety technology has a significant impact on our society. It helps reduce the risk of foodborne illnesses and improve the quality of food products. It can also be used to combat food insecurity by extending the shelf life of fresh food, reducing spoilage, and minimizing food waste. Currently, there is a rapidly growing demand by consumers and food producers for smart food safety solutions that can (i) detect, sense and record deterioration inside the food package and (ii) ensure the safety, preserve quality, and warn about possible problems during food transport and storage. Meat and livestock contribute to personal health and well-being, maintain ecological balances and secure socioeconomic livelihoods. Meat production and meat consumption are frequent subjects of societal debate, and for good reason. As a pivotal source of protein and nourishment and playing a large role in ecological and economic systems, meat production, distribution, and consumption must continuously evolve with the best technologies available to maximize its benefits and minimize undesirable impacts. Meat is a highly perishable food and requires proper storage, processing, and packaging. Meat products decompose naturally because of high fat and water contents that render them susceptible to spoilage by both lipid oxidation and microbial contamination. The spoiled meat is hazardous due to microbial growth and the subsequent transmission of food borne illnesses. Primelabs has been working to develop and commercialize a low-cost add-on system that utilizes a smartphone (Apple iOS or Google Android devices) operating an artificial intelligence (AI) software application (app) and a high-performance multispectral imaging (MSI) camera. Primelabs solution, designed to be used either with either modified atmosphere packaging (MAP) or aerobically packaged meat, aims to monitor the quality rapidly and non-destructively of the meat or food as it awaits to be sold or distributed such that a consumer which could either be an individual person, farmer, meat producer, grocery store, warehouse, hospital, or school can determine whether the meat/food should be repurposed, cooked, given away, or disposed. Changes/Problems:During Phase 1, we proposed to develop a low-cost, clip-based add-on system that utilizes the available camera in a smartphone to create a multispectral imager (MSI). The clip-based add-on system utilized a custom developed 9-position electronically controlled mechanical filter wheel. During Phase I, the cost estimate to custom build the filter wheel increased from our initial estimate of $300 to $900, when done for low volume numbers. The cost of the glass optical filters would cost an additional $600 as our proposed low-cost acrylic color filters did not work as expected. Thus, after carefully reviewing the market needs and technology landscape, we updated our system architecture as follows.During Phase 1, we began to collaborate and formed abusiness partnership with BodkinDesign & Engineering LLC (BDE), which exclusively distributes the Monarch Pro MSI camera from Unispectral in the United States. The Monarch Profrom Unispectral includes a micro-electro-mechanical-system (MEMS) based tunable optical filter.Monarch Pro MSI camera offered better performance than our mechanical filter wheel in terms of form-factor, system response time, spectral band selection, and avoiding cumbersome and unreliable mechanical operation. Further, the use of a dedicated external MSI reduced the variability with using smartphone cameras and made our system compatible with any smart device including Apple's and Google's smartphones and tablets. Further, working with BDE, a larger and more established company, is helping us with customer discovery and technology transition. What opportunities for training and professional development has the project provided?Dr. Kim Phung, Principal Investigator and Research Scientist, Primelabs leveraged the project to gain experience in artifical intelligence algorithms, cloud computing,andmobile application development and understanding food/meat safety application requirements. Dr. Vinaya Gogineni, Sr.Research Scientist, Primelabs leveraged the project to gain experience in data analytics and large database management. Dr. Yvonne Sun, Associate Professor, Department of Biology, University of Dayton, and her graduate student worked on assessing bacterial contamintion of meat samples, preparation of meat samples forquality testing, and meat spoliage dynamics. How have the results been disseminated to communities of interest?Primelabs and Bodkin Design & Engineering (BDE) have formed a business partnership to combinedly offer the smartphone MSI system for meat/food safety area at a price of $3,300 per unit. Working with BDE, a larger and more established company, is helping us with customer discovery and technology transition. The partnership with BDE will help us leverage the extensive customer database of BDE. In addition, we plan to add more customers from supermarkets, meat producers, school districts, senior living facilities, casinos, restaurants, and hospitals. During Phase I, our team approached meat producers, supermarkets chains, restaurants, senior living facilities, schools, and small grocery shops including Whole Foods, Kroger Supermarkets, Dorothy Lane Market and Fresh Thyme. We received several letters of support. As the technology matures, we will expand to various other potential customers and vertical markets. Aside from food supply management, smartphone-MSI systems have multiple applications in pharmaceutical, chemical/biological agent detection for national security, and pharmaceutical fields. Our technology offers several advantages including, (i) non-destructive inspection that can be implemented for under various situations including high speeds for large product volumes, product shelves, and individual point-of-sale inspections, (ii) meat inspection applications that can include external and internal attributes of food, and (iii) automated online imaging that has great potential for real-time quality monitoring, thereby improving the profitability to manufacturers and supermarkets and reducing the cost of food products for consumers. Food safety technology has a significant impact on our society. It helps to reduce the risk of foodborne illnesses and improve the quality of food products. Our technology can be used to combat food insecurity by extending the shelf life of fresh food, reducing spoilage, and minimizing food waste.Reducing the amount of food waste sent to landfills could not only help ease the impact of climate change but also put food in the mouths of millions of people. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? Primelabs, in collaboration with University of Dayton (UD) and Bodkin Design & Engineering (BDE), has been working to develop and commercialize a low-cost add-on system that utilizes a smartphone (Apple iOS or Google Android devices) operating an artificial intelligence (AI) software application (app) and a high-performance multispectral imaging (MSI) camera. Our smartphone-MSI system is designed to be used either with modified atmosphere packaging (MAP) or aerobically packaged meat to provide information about the quality and safety of meat products. Our SBIR project focuses on the development and commercialization of our intelligent decision support system utilizing MSI information for use under various situations including (i) ground or whole meat, (ii) aerobic or MAP storage, (iii) storage at different temperatures (4, 10, and 15 oC), and (iv) different types of meat (such as beef, mutton, chicken, turkey, and pork). During Phase 1, we collected our data from aerobically stored ground and whole beef and pork samples along with the annotation. In particular, the input to training our AI software algorithm is the meat image that is associated with the output which is the meat quality. The samples were prepared and imaged at Dr. Yvonne Sun's, biosafety level-2 laboratory at the University of Dayton (UD). While the meat samples were stored in different temperature conditions, we collected a large dataset of color and multispectral images of meat. Two forms of antibiotic-free meat, ground and whole, were purchased from a local market, and were prepared within 6 hours into sterile plastic petri dish (100 mm in diameter) and placed into 4°C, 10°C, and 15°C incubators. During Phase 1, the focus of our work was on the development of the AI model for determination of the meat quality.In addition, our AI model includes a degree of confidence measure which would allow the users to undertake their standard inspection as a secondary step to prevent food wastage from false positives. Our user interface is designed to be easy to use such that no technical training with our smartphone-MSI system is required to make an informed decision.Our developed AI model and algorithm for MSI image detection and classification is based on Residual Networks (ResNet)-50, which is a popular Convolutional Neural Network (CNN) type architectures for computationally efficient image processing. Our developedAI algorithm developed during Phase I is the world's first such model to be developed for meat/food safety where a smartphone is used to control and power a MSI camera, and then acquire, process, and classify the images from the MSI camera in real-time. During Phase I, we placed significant efforts to optimize and improve the computational efficiency of our ResNet-50 software algorithm. Our AI model trained on the ground beef data achieved 100% in terms of accuracy. Meanwhile, the model trained on the whole beef data achieved 97% in terms of accuracy. When we trained the model on both ground- and whole- beef, the accuracy rate reached 99%. For the pork, we observed the whole pork trained model achieved the perfect performance (100%) for all metrics. Meanwhile, the ground pork trained and the ground and whole pork trained models, respectively reached 97.98% and 98.43% in terms of accuracy.

Publications