Cotton Foreign Matter Detection Using Hyperspectral Transmittance Imaging

Cotton plays an important role in the U.S. national economy. This commodity can be contaminated by various foreign matter (FM) during harvesting and processing, leading to potential damage to textile products. Current sensing methods can only detect the presence of foreign matter on the surface of cotton, but cannot detect and classify foreign matter that is mixed with and embedded inside the cotton. This research focused on the detection and classification of common foreign matter hidden within the cotton lint by hyperspectral transmittance imaging in the spectral range from 950-1650 nm. Three cultivars of cotton and 10 common types of foreign matter were collected from the field and the foreign matter were sandwiched by two thin cotton lint webs. The transmittance imaging platform was designed and optimized for the best performance of the transmittance mode. After acquiring images of cotton and foreign matter mixture, minimum noise fraction (MNF) rotation was utilized to obtain component images to assist visual detection and mean spectra extraction from a total of 141 bands. Linear discriminant analysis (LDA) and support vector machine (SVM) were performed for classification at the spectral and pixel level, respectively. Over 90% of the accurate classification rate was achieved for the spectral data and about 95% for the pixel classification. The preliminary results demonstrated that it was feasible to detect certain types of foreign matter that was buried within cotton using hyperspectral transmittance imaging.