Detecting Escherichia coli Contamination on Plant Leaf Surfaces Using UV-C Fluorescence Imaging and Deep Learning.

Area of Science:

• Food Science
• Microbiology
• Computer Science

Background:

• Escherichia coli (E. coli) contamination on fruits and vegetables poses significant public health risks.
• Outbreaks linked to produce have occurred nationally, highlighting the need for improved food safety measures.
• Rapid and reliable detection methods for E. coli on produce are essential to prevent contamination.

Purpose of the Study:

• To evaluate the performance of the CSI-D+ system with deep learning for detecting E. coli on citrus and spinach.
• To compare the efficacy of various deep learning models, including YOLO11 variants, EfficientNetB7, and ConvNeXtBase.
• To assess the accuracy and speed of fluorescence-based imaging combined with deep learning for E. coli detection.

Main Methods:

• Inoculated citrus and spinach leaves with eight levels of E. coli contamination (0 to 10^8 CFU/mL).
• Captured fluorescence images of contaminated leaf samples.
• Trained multiple deep learning models (EfficientNetB7, ConvNeXtBase, YOLO11 variants) on image data, utilizing Eigen-CAM for visualization.

Main Results:

• All YOLO11 models outperformed EfficientNetB7 and ConvNeXtBase in E. coli detection.
• YOLO11s-cls demonstrated the highest performance, with average test accuracies of 85.93% (citrus) and 92.00% (spinach).
• The YOLO11s-cls model achieved a fast inference speed of 0.011 seconds per image and a small model size of 11 MB.

Conclusions:

• Fluorescence-based imaging coupled with deep learning offers a rapid and reliable method for E. coli detection on produce.
• The YOLO11s-cls model shows significant potential for real-time food safety applications.
• This technology can aid in timely interventions to prevent contaminated produce from reaching consumers.