Image Processing-Based Detection Method for Corn Damage Ratio and Synchronous Analysis of Quality Variation

Traditional methods for detecting the corn damage ratio (e.g., manual sorting and weighing, and electric sieving) are time-consuming and labor-intensive, failing to meet the demand for real-time quality assessment during storage and transportation. To address this, this study proposed a rapid detection method for the corn damage ratio based on image processing and deep learning. Furthermore, the impact of the damage ratio on corn quality deterioration following long-distance transportation and storage was simultaneously analyzed. This research aimed to provide a technical support and a theoretical basis for intelligent corn quality monitoring, early risk warning, and the optimization of storage and transportation conditions.

A total of 500 raw corn kernels images captured by smartphones were utilized as the dataset. A rapid prediction model for the corn damage ratio was developed using image processing techniques (image rectification, preprocessing, segmentation, and area extraction) combined with the training and evaluation of three deep learning models (Resnet50, Densenet121, and Vision Transformer [ViT]). Additionally, by simulating the temperature and humidity conditions of the waterway and land routes in China’s “North-to-South Grain Transportation” system, the quality variation of corn with different damage ratios (0-12%) following storage and transportation was systematically investigated.

The ViT model demonstrated optimal performance in identifying damaged corn kernels, achieving both accuracy and precision rates of 99%. The mean absolute error between the predicted and actual values was merely 0.45%, with a coefficient of determination (R²) reaching 0.978. As the damage ratio increased from 0 to 12%, the physicochemical properties of corn transported via waterway and land routes showed significant increases (P<0.05): moisture content rose by 1.384% and 0.461%, fatty acid values increased by 7.92 mg KOH/100 g and 4.49 mg KOH/100 g, electrical conductivity elevated by 6.72 and 4.66 μs·cm^-1, and malondialdehyde (MDA) content increased by 38.73% and 19.22%, respectively. Regarding nutritional quality, the protein content decreased, while the starch content exhibited a fluctuating trend of an initial increase, followed by a decrease, and a subsequent rise. An imbalance emerged between the amylose (AM) and amylopectin (AP) content, accompanied by significant alterations in the pasting properties.

This study provided a reliable technical approach for the rapid detection of the corn damage ratio, offering a scientific basis for optimizing storage and transportation conditions to mitigate quality deterioration. Specifically, it was recommend establishing a 4% damage ratio as the critical control threshold in long-distance and high-humidity storage and transportation scenarios, such as China’s “North-to-South Grain Transportation” system. Batches exceeding this threshold should undergo priority sorting or quality preservation treatments.