Identification of Cotton Foreign Matter Using Line Scan Hyperspectral Transmittance Imaging

Cotton foreign matter (FM) is detrimental to fiber quality as it may damage cotton fiber during ginning processing or cause flaws in finished textiles. Therefore, detecting and classifying foreign matter are very important steps in the cotton production process. The aim of this study was to identify various types of cotton foreign matter embedded within lint webs using hyperspectral transmittance imaging at the spectral range of 400–1000 nm. A total of 11 types of foreign matter and five cultivars of cotton were collected from the field and the foreign matter was placed between two thin lint webs. A push-broom based hyperspectral imaging system was used to acquire images in transmittance mode. The acquired hyperspectral images were corrected using flat field correction and cropped due to noise at the edges. The foreign matter was observable in grayscale images. The image at 500 nm was chosen to be used for manual region-of-interest (ROI) selection. Mean transmittance spectra were extracted from the ROIs and normalized across all samples. Canonical discriminant analysis (CDA) was used to group FM, and multivariate analysis of variance (MANOVA) was employed to evaluate the differences between each combination of two types of FM using the first three canonical variables. The support vector machine (SVM) classifier was used to classify FM and was verified by 3-fold cross-validation. The classification results indicated that it is feasible to identify FM using this method since plastic package, paper, seed meat, and green leaf were well classified. The accuracies of distinguishing types of FM which have similar appearance and similar chemical content were lower but exceeded 60%. The average classification rate of all types of FM and cotton lint was 79.2%.