It is desirable to predict lint cotton color in advance of processing
the cotton in a gin. Improvements over the use of seed cotton color as
the lone predictor are needed. Standard lint color and trash content
measurements were made on 61 samples of lint and seed cotton to determ
ine the predictability of lint color from that of seed cotton. Also, v
isible spectral data were collected from the seed cotton and lint samp
les, and from corresponding samples of fuzzy and delinted seed. Simple
and multiple linear regression were conducted to determine the relati
onships among lint color, seed cotton color, and spectral data. Trash
content data and spectral data, from cotton seed were applied in addit
ion to seed cotton data in an effort to enhance the predictability of
lint color. Results from linear regression showed that seed cotton col
or correlated moderately (R(2) approximate to 0.6) with lint color. Se
ed cotton and lint reflectances at individual 50-nm spectral bands cor
related poorly (R(2) approximate to 0.2). With trash content in the an
alysis, the fit was improved (R(2) approximate to 0.4). Seed spectral
data also improved the correlations (R(2) approximate to 0.4). In addi
tion, seed spectral data improved the correlations between seed cotton
color and lint color (from R(2) approximate to 0.6 to R(2) approximat
e to 0.7). Ratios of seed spectral data were about as effective as the
spectral data themselves, The inclusion of seed cotton spectral data
in these models improved correlations slightly more (from R(2) approxi
mate to 0.7 to R(2) approximate to 0.8). Adjusting lint and seed cotto
n spectral data for trash and seed reflectances was largely unsuccessf
ul in improving correlations between lint and seed cotton spectral dat
a. The regression methods and data relationships mentioned above are d
iscussed in detail in the article.