Multispectral Sensor for In-Situ Cotton Fiber Quality Measurement

Abstract

Reflectance spectra of cotton fiber samples having different fiber quality levels were measured with a high-resolution spectrophotometer, processed with waveband averaging and wavelet analysis, and then related to micronaire with multiple linear regression. Regression models indicated that micronaire had a close relationship (R2 = 0.89) with reflectance at seven 100 nm wavebands (1120, 1296, 1550, 1664, 1852, 2020, and 2340 nm). In the wavelet-based analysis, six wavelet-coefficient regressors were identified and entered into a regression model. This model also indicated a very close relationship between micronaire and reflectance spectra (R2 = 0.97). A prototype cotton fiber quality sensor was developed based on the characteristics of the cotton fiber reflectance spectrum and the wavelet-based multiple-regression analyses. The sensor consists of a VisGaAs camera, optical bandpass filters, a halogen light source, and an image collection and processing system. Images of lint samples at three near-infrared (NIR) wavebands (1450, 1550, and 1600 nm) were acquired and analyzed with two methods to determine the relationship between image pixel values and cotton fiber micronaire. One method involved ROI (region-of-interest) pixel-value data, while the other involved histogram-based pixel-value data. Results showed that the sensor was capable of accurately estimating the fiber micronaire (R2 = 0.99 for ROI data, and R2 = 0.99 for histogram-based data). This sensor could potentially be used for measuring cotton fiber quality along with spatial data from a GPS receiver as the cotton is harvested in the field, making it possible to generate cotton fiber quality maps. The sensor also has the potential to be used for segregating cotton at harvest based on fiber quality.