Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/2/2018
Publication Date: 4/1/2018
Citation: Wu, T., Armstrong, P.R., Maghirang, E.B. 2018. Vis- and NIR-based instruments for detection of black-tip damaged wheat kernels: A comparative study. Transactions of the ASABE. 61(2):461-467. https://doi.org/10.13031/trans.12432.
DOI: https://doi.org/10.13031/trans.12432 Interpretive Summary: Black-tip (BT) present in wheat kernels is a non-mycotoxic fungus that causes the formation of dark brown or black sooty mold at the tip of the wheat kernel. This condition has a negative economic impact at marketing and black tip damaged wheat is limited to two percent for U.S. No. 1 wheat. Three instruments using the measurement of reflected light from the kernel were examined as a method to detect black-spot on single kernels. The instruments used wavelengths that were entirely visible, visible plus near-infrared or entirely near infrared. When evaluating kernels with differing levels of black-tip (sound or no BT, low or high black-tip symptom (BTS), and black-tip damaged (BTD)), classification of black tip severity was best when sound and low BT kernels were combined and compared against combined BT symptom and damaged kernels (BTD). The instrument using the combination of visible plus near-infrared light performed best (86% accuracy) compared to the entirely visible (76% accuracy) or entirely near-infrared instruments (78% accuracy). Two other combinations of damage level categories were evaluated but were not classified as well as the combination above. These combinations were between 1) all four kernel levels of sound, low BT, high black-tip symptom (BTS) and black-tip damaged (BTD); 2) sound, low BT combined with high black-tip symptom (BTS), and black-tip damaged kernels (BTD). This work shows that visible and near-infrared reflectance spectroscopy could be useful as a quality screening tool and is a method that plant breeders could use when breeding for resistance to black tip formation
Technical Abstract: Black-tip (BT) present in wheat kernels is a non-mycotoxic fungus that attacks the kernels wherein any of a number of molds forms a dark brown or black sooty mold at the tip of the wheat kernel. Three spectrometers covering the spectral ranges 950-1636nm (Spec1), 600-1045nm (Spec2), and 380-780nm (Spec3) were evaluated for the ability to predict the presence of black-tip. Kernels with different levels of black-tip damage (BTD) were quantified into four levels: sound (A), low (B) or high (C) black-tip symptom (BTS), and black-tip damaged (D). Discriminant classification models were developed to evaluate combinations of levels. The combinations were 1) A, B, C and D separately; 2) A, B+C, and D; and 3) A+B and C+D. The spectral data for 2,760 kernels obtained from 23 hard red winter wheat samples composing of visually selected thirty kernels for each of the four levels of black-tip severity (A, B, C, and D) were collected with each spectrometer. Discriminant calibration models for each spectrometer and classification category were developed based on (a) three combinations of 17 samples with the six remaining samples used as independent validation, and (b) combinations of 20 randomly-picked kernels from each of the 23 HRW samples as calibration samples with the remaining ten kernels used as validation samples. Discriminant analysis was based on five wavelengths for each model. Spectra pretreatment was the standard normal variate (SNV). Results showed that all three spectrometers were capable of detecting BTD on wheat kernels. BT classification accuracy was observed to have been affected by wheat varieties for Spec1 and Spec2 (both with NIR wavelengths) but not for Spec3, which was purely in the visible region. The 2-category classification (A+B, C+D) provided higher accuracy than 3-category (A, B+C, D) and 4-category classifications (A, B, C, D). Based on percent correct classification and Youden’s index, Spec2 performed better for detecting sound and BT wheat kernels with classification accuracies of the best 2-category classification calibration model ranging from 85.6 to 87.5% compared to Spec1 at 74.8 to 78.4% and Spec3 at 76.7 to 79.2%.