Practicability investigation of using near-infrared hyperspectral imaging to detect rice kernels infected with rice false smut in different conditions

• Published in: Sensors and actuators. B, Chemical Chemical, 2020-04, Vol.308, p.127696, Article 127696
• Main Authors: Wu, Na, Jiang, Hubiao, Bao, Yidan, Zhang, Chu, Zhang, Jingze, Song, Wenjian, Zhao, Yiying, Mi, Chunxiao, He, Yong, Liu, Fei
• Format: Article
• Language: English
• Subjects: Hyperspectral imaging; Pathological analysis; Practicability analysis; Rice false smut (RFS); Rice kernels
• ISSN: 0925-4005; 1873-3077
• DOI: 10.1016/j.snb.2020.127696

[Display omitted] •Detecting rice kernels with different varieties, different infection conditions and different infection status;•Combining microscopic molecular detection technology with macroscopic spectral imaging technology;•Investigating the practicality and generalization of the detection model;•Boosting large-scale seeds detection in modern seed industry. Rice false smut (RFS) is a devastating seed-brone rice disease in many rice-growing countries, endangering the health of rice germplasm resources and reducing the yield and quality of rice. This study aimed to propose an effective method for RFS detection in actual production based on near-infrared hyperspectral imaging (NIR-HSI) paired with pathological analysis. The true infection status of rice kernels collected in different conditions was labeled by PCR. The separability between healthy and infected rice kernels was explored using principal component analysis (PCA). Multivariate quantitative analysis models were constructed based on full wavelengths of laboratory-inoculated kernels. Characteristic wavelengths extracted to improve detection performance contained fingerprint information related to RFS infection. The best classification accuracies for healthy and infected mixed kernels with different infection degrees achieved 99.33 % on calibration set and 99.20 % on prediction set, respectively, using RF-ELM model. The practicality of detection model was further verified through obtaining detection accuracies of 91.07 % and 89.38 % for two varieties of field-infected rice kernels and visualizing the category attribute of single rice kernel in hyperspectral images. The overall results indicated the excellent potential of NIR-HSI for on-line large-scale seeds detection in modern seed industry.