Skip to main content
ARS Home » Research » Publications at this Location » Publication #274455
Title: Reference and NIR model: an example of trash measurements in cotton fibers

Author
item Liu, Yongliang
item Thibodeaux, Devron
item Gamble, Gary

Submitted to: NIR news (Near Infrared Reflectance News)
Publication Type: Popular Publication
Publication Acceptance Date: 11/14/2011
Publication Date: 12/1/2011
Citation: Liu, Y., Thibodeaux, D.P., Gamble, G.R. 2011. Reference and NIR model: an example of trash measurements in cotton fibers. NIR news (Near Infrared Reflectance News). 22(8):6-8.

Interpretive Summary: Reliable and precise reference determination is one of most crucial factors that influence the NIR model efficiency. In some scenarios, it is a great challenge to acquire accurate and consistent references of physical properties, especially if the samples are highly heterogeneous and are not easy to be processed (such as being ground) for their uniformity. In an earlier report, we proposed a pre-screening procedure to determine appropriate calibration samples on the basis of two independent measurements of cotton fiber strength, prior to NIR model development. In turn, we applied the same strategy to trash components in cotton fibers, and undoubtedly, it led to removal of some samples (or outliers) from the consideration. Ultimately, this would generate one interesting point, i.e., what is NIR spectral response to those outliers. Therefore, this Note examines the NIR model performance on cotton trash contents for the samples belonging to various classes. The outcome provides NIR researchers a new sight for rapid, accurate, and routine application of NIR technique.

Technical Abstract: Despite a common trend in describing trash amounts between two tests (HVI vs. SA), the samples were sub-grouped subjectively according to their ratio valuess. From the least intercept of respective plot, it was proposed the relationship of SAvisible = 6.82*HVIarea. In order to verify this, NIR spectra were correlated with HVIarea readings. The redeveloped models highlight different response to validation samples in various subsets. Of interest is that the model from samples belonging to the identified relationship suggests the highest r2, RPD, and means/SEP, likely demonstrating the better determined references for the samples in this subset than those in others.