Evaluating cotton length uniformity through comprehensive length attributes measured by dual-beard fibrography

Abstract

The quality of cotton yarn, such as evenness and strength, relies on not only the overall length of spun fibers but also fiber length uniformity. In the current cotton classification system (Cotton Incorporated in Classification of upland cotton, 2018; USDA in The classification of cotton, Agricultural Marketing Service, Washington, DC, 1995), length uniformity is measured by a single factor—uniformity index (UI), which does not explicitly include short fiber content (SFC) and neglects the interactive effects among length attributes. The goal of this study was to search for key length attributes and new classification methods for more comprehensive evaluations of cotton length uniformity. We firstly investigated the associations of length attributes measurable by the dual-beard fibrography (DBF) (Zhou et al. in Text Res J 90(1):37–48, 2020) to select a set of key features to reduce the dimensionality for consecutive statistical analysis. This set contains an overall length attribute (upper half mean length—UHML), SFC and UI that represent more realistic information about cotton quality. We then used the K-means clustering to determine the natural clusters of the length uniformity based on the data of 29 selected cotton samples that have a wide range of fiber length distributions. The clustering resulted in six optimal clusters, each representing a group of homogeneous length attributes. Thirdly, we adopted one support-vector-machine (SVM) classifier for cotton length uniformity prediction on unknown fibers. To verify the prediction accuracy, 25 new specimens were taken from the 29 samples used in the K-mean clustering to run the DBF test and the SVM classification. It was found that 92% of these specimens yielded the same cluster numbers as the ones resulted from the clustering. In summary, UHML, SFC and UI represent more comprehensive length attributes of cotton, and the six new clusters from the K-mean clustering offer more holistic evaluation on cotton length uniformity.