Mildew detection in rice grains based on computer vision and the YOLO convolutional neural network

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Traditional rice microbial detection methods (culture, sensory) are slow and require expertise.
• These methods are unsuitable for rapid, on-site testing during rice storage or trade.
• A need exists for fast, non-destructive techniques for rice quality assessment.

Purpose of the Study:

• To develop a rapid, non-destructive method for detecting rice mildew.
• To utilize micro-computer vision and a YOLO-V5 convolutional neural network for mildew detection.
• To establish a correlation between detected mold coverage area and bacterial colony counts.

Main Methods:

• Collected images of mildewed rice samples from multiple locations.
• Employed a YOLO-V5 convolutional neural network model for detecting and quantifying moldy areas.
• Correlated the estimated mold coverage area with the total bacterial colony count.

Main Results:

• The YOLO-V5 model achieved 82.1% precision and 86.5% recall in identifying mildewed rice areas.
• For light mold detection, precision reached 100% and recall was 95.3% based on mean mildewed area.
• The computer vision and YOLO-V5 approach demonstrated high accuracy in mildew detection.

Conclusions:

• Micro-computer vision combined with the YOLO convolutional neural network offers a viable solution for rapid rice mildew detection.
• This method is suitable for real-time quality assessment of rice during storage and trade.
• The technology facilitates efficient and accurate identification of mold contamination in rice.