Cotton Fiber Quality Characterization with Vis-NIR Reflectance Spectroscopy: Toward An Optimal Sensor

The objectives of this research were to (1) develop calibration models to predict High Volume Instrument (HVI) fiber quality parameters with Vis-NIR reflectance of lint cotton, and (2) compare different methods for Vis-NIR model calibration. This study is directed toward the development of opto-electronic sensors to measure cotton fiber quality in real time in situ. Sixty seed cotton samples of two varieties were handpicked and ginned with a laboratory-scale saw-type gin. Ginned lint samples were measured with a Cary 500 UV-Vis-NIR spectrophotometer in a wavelength range from 400 to 2500 nm. A portion of the lint samples was subjected to HVI measurement of six fiber quality properties: micronaire, length, strength, uniformity, brightness (Rd), and yellowness (+b). Two methods, band-averaging (BA) and discrete wavelet transform (DWT), were used to preprocess the lint spectra. Calibration models for each fiber quality property were developed with partial least squares regression (PLSR) and multiple linear regression (MLR); and the performance of the models was assessed with leave-one-out cross validation and root mean squared error (RMSE). Among all six fiber quality properties, micronaire and +b can be most successfully predicted, with R² greater than 0.80 and 0.75, respectively. Prediction of length, uniformity, and strength was moderately successful, with R² ranging from 0.55 to 0.70. Prediction of Rd was poor (R² < 0.4). More interestingly, the RMSE of the calibration model for most fiber quality properties approaches the measurement accuracy of HVI instruments specified by USDA-AMS, indicating that a Vis-NIR opto-electronic sensor could perform with an acceptable level of measurement accuracy. Among the different model calibration methods, DWT-MLR generally performed better than BA-MLR and BA-PLSR, which can be attributed to the fact that DWT considers wavebands of varying bandwidths for incorporation into the model. Development of opto-electronic sensors based on Vis-NIR reflectance of cotton lint appears promising.