Study on Rice Grain Mildewed Region Recognition Based on Microscopic Computer Vision and YOLO-v5 Model

Area of Science:

• Agricultural Science
• Food Science
• Computer Vision

Background:

• Mold contamination in rice poses significant food safety and quality risks.
• Current detection methods can be slow, destructive, or lack precision.
• Accurate quantification of mold is crucial for quality control and trade.

Purpose of the Study:

• To develop a high-speed, nondestructive method for detecting mildewed rice grains.
• To utilize advanced computer vision techniques for mold identification and quantification.
• To establish a correlation between visible mold area and microbial load.

Main Methods:

• Acquisition of microscopic images of rice grains contaminated with Aspergillus niger, Penicillium citrinum, and Aspergillus cinerea.
• Development and application of three YOLO-v5 models for identifying mildewed regions in microscopic images.
• Analysis of the relationship between the proportion of mildewed area and the total viable count (TVC) of fungi.

Main Results:

• YOLO-v5 models achieved high detection accuracy: 89.26% for A. niger, 91.15% for P. citrinum, and 90.19% for A. cinerea.
• A logarithmic correlation was found between the mildewed region area and the logarithm of the total viable count (TVC).
• Determination coefficients for the correlations were 0.7466 (A. niger), 0.7587 (P. citrinum), and 0.8148 (A. cinerea).

Conclusions:

• The developed YOLO-v5 based method offers a rapid and nondestructive approach for mildewed rice detection.
• The correlation analysis provides a quantitative link between visual mold detection and microbial contamination levels.
• This research serves as a foundation for advanced mildewed rice detection using machine vision technology.