Correlation of NIR Data with Cotton Quality Characteristics

A desire for improvement of the state of the art in cotton quality measurements, and for application of such measurements to the ginning process, has prompted investigation of near-infrared reflectance (NIR) as a means of measuring different cotton quality characteristics. Cotton samples from five different cotton sources covering a range of white color grades were ginned with three different lint-cleaner treatments. The lint samples then underwent a full set of High Volume Instrument quality tests at a cotton classing office. The samples were subsequently measured on an NIR spectrophotometer at 2-nm intervals from 1100 to 2500 nm. These 700 measurements were averaged into 14 bands of 100 nm in width. Ratios of the bands were calculated, and statistical model selection procedures were used to establish optimal regression models for the quality characteristics with continuous values. Correlations of the models to the quality characteristics were also calculated. Optimal models showed the following R2 values in relation to their corresponding cotton quality characteristics0.88 for reflectance, 0.85 for yellowness, 0.60 for trash content, 0.96 for micronaire, 0.73 for strength, 0.79 for length, and 0.67 for length uniformity. The NIR ratios were also employed in discriminant analyses to establish models for classification of discrete attributes of the cotton samples. Classification accuracies were 93% for cotton source, 91% for combined source and lint-cleaner treatment, 93% for color grade, and 70% for color-grade quadrant.