AI Deep-learning classifies yeast spoilage in record time.
Could this be the future of improving food shelf life and reducing food-borne illnesses?
The Problem with Yeast
Yeast is an excellent microorganism used for fermenting foods into beer, wine and cheese. However, unwanted yeasts can also spoil dairy products, fruit juices, and more. Currently, yeast identification methods take a long time to process, but many foods are already on the market before then. To combat these problems, UC Davis researchers Hyeon Woo (Howard) Park, J. Mason Earles, and Nitin Nitin experimented with deep learning to create a yeast classification model that can provide accurate results in 1/20 of the typical time. “For yeast and fungi, it typically requires 3-5 days of culture,” Dr. Nitin explains. “This method is able to answer the same question in 6 to 7 hours.”
The Experiment
Seven yeast species were utilized because of their prevalence in the food industry:
- C. albicans
- S. cerevisiae
- G. candidum
- R. babjevae
- Y. lipolytica
- D. hansenii
- W. anomalus
The bacteria were put on agar plates for incubation in increments of three to seven hours. After incubating, the bacteria were photographed using white light microscopy with a 20X objective. 300 images were collected over three experiments.
To enhance the datasets, a model made for generating synthetic images of yeast called GAN was used to make 400 more images for each species. These synthetic images were evaluated and the classification model was retrained with both original and synthetic datasets. “There are some colonies that look alike, so it was a challenge to discriminate between the two microcolonies,” Howard explained. “So that’s why we tried adopting the GAN model…to generate synthetic images that mimic realistic images.”
The Final Test
After training the deep-learning model, it was time to put it to the test. Store-bought tomatoes and tomato juice were infected with the seven different species and plated for incubation. After six hours, the microbial samples were imaged and evaluated using the AI model. The yeast classification model was accurate, with an overall precision of 96%.
What Does This Mean for the Future of Food?
Since this research was inspired by industry needs, it was already designed to be quickly implemented. “These AI models are used for object detection in many applications, from cars to other objects on the road. In this study, these models were fine-tuned to detect specific yeast microcolonies on culture plates,” Nitin responds. As for the next steps in detection, Howard explains they are “working on developing AI models and their applications for parasite detection, antibiotic resistance among bacteria,” and the other food safety-related applications. This is only a glimpse into what AI can do for microbial detection, and how this technology can bring next-level protection for growers, processors and eventually to consumers like you!