Skip to main content
ARS Home » Research » Publications at this Location » Publication #103280


item Archibald, Douglas
item Akin, Danny

Submitted to: Near Infrared Spectroscopy International Conference Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: 7/29/1999
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Rapid, simple, and repeatable methods are needed at two key stages of flax processing: (1) during microbial or enzymatic 'retting', which aims to release the fibers from the stem matrix, and (2) after mechanical processing to extract the fibers. NIR reflectance spectroscopy of intact stems over the range 1100 - 2500 nm was found to be suitable for predicting gthe degree of enzymatic retting in the presence of interference due to moisture content and the stem packing geometry. Multivariate regression analysis identified several spectral windows that could effectively monitor retting with only three or four regression factors. A related study aims to measure qualities of flax fiber by NIR. Reflectance spectra of 68 fiber samples from various sources were measured with multiple packs and humidity treatments. The strategy is to remove the repack and humidity variance with spectral preprocessing prior to regression and classification studies. .A preliminary evaluation of the data used unsupervised principal component analysis and category variables to assign the variance in normalized second-derivative spectra of an untreated set of samples. The first factor, summarizing 54% of the spectral variance, is primarily C-H signal that differentiates samples processed by different retting procedures (water, dew and enzyme). Factor 2 (22%) has a strong contribution from moisture bands. Factors 3 (14%), 4 (3%) and 6 (1%) are associated with pigmentation, cultivar, and trash content, respectively. Factor 5 (2%) and other factors (4%) have not been assigned. Reference data for fiber strength and fineness are available for a small subset of fibers (n=15); two-factor PLS regression models show some ability to predict these parameters: strength (r=0.69) and fineness (r=0.83).