Related Experiment Video
Updated: Jun 19, 2026
10:37
Single-cell Microfluidic Analysis of Bacillus subtilis
Published on: January 26, 2018
Rapid Screening of Spreading-Type Bacillus Using a YOLO-Based Classifier on Sheet Media
Aiho Kushida1, Rin Tsuchiya1, Tatsuya Tominaga2
1Division of Food and Nutrition, Graduate School of Human Sciences and Design, Japan Women's University, 2-8-1, Mejirodai, Bunkyo-ku, Tokyo 112-8681, Japan.
Journal of Food Protection
|June 17, 2026
Summary
A new deep learning classifier accurately distinguishes and counts aerobic spore-forming bacteria colonies on food safety media. This AI tool aids in rapid food quality assessment and safety management for users of all skill levels.
Area of Science:
- Microbiology
- Food Science
- Artificial Intelligence
Background:
- Aerobic spore-forming bacteria are significant contributors to food spoilage and foodborne illnesses.
- Distinguishing these bacteria from others typically relies on visual colony morphology on specialized media.
- Accurate and efficient quantification is crucial for food safety and quality control.
Purpose of the Study:
- To develop and validate a deep learning-based classifier for automated detection and enumeration of aerobic spore-forming bacterial colonies.
- To assess the accuracy, versatility, and practical feasibility of the developed classifier in food safety applications.
- To provide a rapid, objective method for quantifying spore-forming bacteria, complementing traditional methods.
Main Methods:
- Development of a classifier using the YOLO (You Only Look Once) deep learning algorithm.
- Training the classifier with images of Bacillus subtilis subsp. spizizenii (spore-forming) and Escherichia coli (non-spore-forming) colonies.
- Evaluation of classifier performance on 60 images (4549 colonies) and testing its versatility across 46 strains from 34 species.
- Assessing practical feasibility using 66 ready-to-eat food samples and comparing results with 16S rDNA sequencing.
Main Results:
- The YOLO-based classifier achieved an accuracy exceeding 0.98 in initial performance evaluations.
- Mean accuracy greater than 0.94 was observed when tested for versatility across diverse bacterial species.
- The classifier demonstrated a precision approaching 0.91 when applied to ready-to-eat food samples, with rapid enumeration at approximately 133 msec per medium.
- Discrimination power was found to be equivalent to manual visual inspection.
Conclusions:
- The developed deep learning classifier effectively distinguishes and quantifies aerobic spore-forming bacteria colonies with high accuracy and speed.
- This AI tool offers a practical, efficient, and objective solution for routine food safety monitoring and quality management.
- The classifier's performance suggests its potential to assist users of varying expertise in rapidly assessing food safety and quality.
