Data Analysis and Feature Selection Techniques for a Machine Learning Process to Assess Maturity in Cotton Fibers
As world leaders in cotton production, the United States contributes to more than 40% of the world's total exports. Cotton is primarily used by the textile industry for manufacturing products such as clothing and home furnishings, which demand higher quality fibers. Among the physical characteristics of cotton fibers, maturity is one of the most important. In a prior work, a system was devised to measure maturity using longitudinal images of individual cotton fibers. This project will add a feature selection strategy that uses a transfer learning process to evaluate the suitability of a strategically selected subset of target domain features. The source domain is defined as the two features of interest from the current reference methoda cross-sectional view of a cotton fiber measured with a microscope. The target domain is defined as the thirteen features nine standard features and four Haralick texture features measured from longitudinal view scans by a line-scan camera. The adapted framework provides an optimal subset of target features to best assess the maturity of cotton fibers.