Skip to main content
ARS Home » Southeast Area » Tifton, Georgia » Crop Genetics and Breeding Research » Research » Publications at this Location » Publication #380714

Research Project: Genetic Improvement of Maize and Sorghum for Resistance to Biotic and Abiotic Stresses

Location: Crop Genetics and Breeding Research

Title: Classifying Maize Kernels Naturally Infected by Fungi Using Near-infrared Hyperspectral Hmaging

item CHU, XUAN - Zhongkai University
item WANG, WEI - China Agricultural University
item Ni, Xinzhi
item LI, CHUNYANG - Jiangsu Academy Agricultural Sciences
item LI, YUFENG - Nanjing Forestry University

Submitted to: Infrared Physics and Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/15/2020
Publication Date: 2/17/2020
Citation: Chu, X., Wang, W., Ni, X., Li, C., Li, Y. 2020. Classifying Maize Kernels Naturally Infected by Fungi Using Near-infrared Hyperspectral Hmaging. Infrared Physics and Technology. 105: Article 103242.

Interpretive Summary: Corn is a staple cereal with high nutritional value and characteristic flavor. Corn production was 1.1 billion tons in 2017 worldwide. It is used for human consumption in diverse forms, and plays an important role in animal feed as well. Due to its rich nutrient values and large embryos, maize is easily to be infected by fungi in field, and further exacerbated under warm and humid post-harvest conditions. The main diagnostic tests at present are focused on molecular and analytical methods that have high detection accuracy and sensitivity. The main limitations of those techniques are destructive to samples, high cost, labor intensive, and require the use of unfriendly chemicals. Therefore, non-destructive and rapid detection techniques for fungal infection and mycotoxin contamination have become a major area of interest in recent years. This study was primarily aimed at evaluating the feasibility of utilizing near-infrared hyper-spectral imaging technology with two data extraction strategies, i.e., pixel-wise and object-wise, in classifying maize kernels naturally infected by fungi that causes ear rot. This work indicated that hyperspectral imaging is a useful tool to identify maize kernels naturally infected with fungal pathogens. The finding provides the much-needed baseline information for further research on possibly removal of fungus-infected and mycotoxin-contaminated kernels at harvest, as well as at the grain processing facilities at postharvest.

Technical Abstract: Maize is easily to be infected with fungal pathogens in the 'eld, which causes a considerable yield reduction and mycotoxin contamination under the warm climate. The objective was to assess, with the use of a near-infrared hyperspectral imaging (900–1700 nm), the di'erence between healthy and fungus-infected maize kernels of three hybrids (i.e., ‘JingKe968’, ‘JingNuo2000’ and ‘XianYu335’, representing dent, waxy, and semi-'int endosperms, respectively). Two sampling strategies, i.e., pixel-wise (PW) and object-wise (OW), were utilized to examine the samples and then further analyzed by using principal component analysis (PCA), successive projections algorithm (SPA) and support vector machine (SVM) modelling. In object-wise analysis, average spectra of whole individual kernel were examined. OW-PCA-SVM models developed using PC1 and PC3 through PC6 achieved accuracies of 99.00%, 97.96% and 97.87% for the three maize hybrids respectively. Eight (1168, 1344, 1414, 1428, 1520, 1269, 1691 and 1205 nm), eight (1161, 1336, 1450, 1638, 1266, 1698, 1189 and 1520 nm), and ten (1168, 1375, 1673, 1602, 1340, 1189, 1245, 1534, 1417 and 1262 nm) optimal wavelengths were respectively selected by SPA, and similar results (100%, 98.98% and 98.94%) were achieved. In pixel-wise analysis, classi'cation accuracies at kernel level for the three varieties were 92.00%, 95.3% and 99.32% for PW-PCA-SVM models, and 100%, 100% and 100% for PW-SPA-SVM models, respectively. The results indicated that both object-wise and pixel-wise methods can be used for classi'cation of fungus-infected maize kernels. Pixel-wise classi'cation was superior in generating visualization maps for presenting the spatial information of the fungus-infected maize kernels.