Shortwave infrared hyperspectral Imaging for cotton foreign matter classification

Authors: ZHANG, RUOYU - Shihezi University; LI, CHANGYING - University Of Georgia; ZHANG, MENGYUN - University Of Georgia; Rodgers Iii, James

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/20/2016
Publication Date: 6/27/2016
Citation: Zhang, R., Li, C., Zhang, M., Rodgers III, J.E. 2016. Shortwave infrared hyperspectral Imaging for cotton foreign matter classification. Computers and Electronics in Agriculture. p. 260-270. doi:1016/j.compag.2016.06.023.

Interpretive Summary: Various types of cotton foreign matter seriously reduce the commercial value of cotton lint and further degrade the quality of textile products for consumers. This research was aimed to investigate the potential of a non-contact technique, i.e., liquid crystal tunable filter (LCTF) hyperspectral imaging, for inspection of the foreign matter on the cotton lint surface. The foreign matter samples used in this study included 11 botanical foreign matter (i.e., stem inner, stem outer, bark inner, bark outer, brown leaf, green leaf, bract, hull, seed coat inner, seed coat outer, and seed meat) and 5 non-botanical foreign matter (i.e., twine, module cover, plastic mulch, drip irrigation belt, and poly woven bag). Hyperspectral images of the foreign matter on top of the cotton lint surface were acquired from 480 samples (30 for each type of the foreign matter) using a LCTF hyperspectral imaging system with a spectral range from 900 to 1700 nm. To keep the thickness of the cotton lint web uniform, a piece of borosilicate glass was used to cover the cotton lint and foreign matter samples during the imaging acquisition. The mean spectra of the foreign matter and lint samples were extracted from the regions of interest manually. Stepwise linear discriminant analysis was applied to select the key wavelengths and to discriminate various foreign matter and cotton lint according to their spectral features. The results showed that the borosilicate glass slide did not have a significant effect on the spectral properties of the FM and lint samples. Overall, using near infrared hyperspectral imaging, 96.5% and 95.1% were correctly classified in the leave-one-out and four-fold cross-validation, respectively. Meanwhile, various foreign matter materials could be discriminated by the support vector machine using top minimum noise fraction components transformed from the image. The results demonstrated that the non-contact liquid crystal tunable filter hyperspectral imaging technique is a very effective method to discriminate various foreign matter materials from cotton lint.

Technical Abstract: Various types of cotton foreign matter seriously reduce the commercial value of cotton lint and further degrade the quality of textile products for consumers. This research was aimed to investigate the potential of a non-contact technique, i.e., liquid crystal tunable filter (LCTF) hyperspectral imaging, for inspection of the foreign matter on the cotton lint surface. The foreign matter samples used in this study included 11 botanical foreign matter (i.e., stem inner, stem outer, bark inner, bark outer, brown leaf, green leaf, bract, hull, seed coat inner, seed coat outer, and seed meat) and 5 non-botanical foreign matter (i.e., twine, module cover, plastic mulch, drip irrigation belt, and poly woven bag). Hyperspectral images of the foreign matter on top of the cotton lint surface were acquired from 480 samples (30 for each type of the foreign matter) using a LCTF hyperspectral imaging system with a spectral range from 900 to 1700 nm. To keep the thickness of the cotton lint web uniform, a piece of borosilicate glass was used to cover the cotton lint and foreign matter samples during the imaging acquisition. The mean spectra of the foreign matter and lint samples were extracted from the regions of interest manually. Stepwise linear discriminant analysis was applied to select the key wavelengths and to discriminate various foreign matter and cotton lint according to their spectral features. The results showed that the borosilicate glass slide did not have a significant effect on the spectral properties of the FM and lint samples. Overall, using near infrared hyperspectral imaging, 96.5% and 95.1% were correctly classified in the leave-one-out and four-fold cross-validation, respectively. Meanwhile, various foreign matter materials could be discriminated by the support vector machine using top minimum noise fraction components transformed from the image. The results demonstrated that the non-contact liquid crystal tunable filter hyperspectral imaging technique is a very effective method to discriminate various foreign matter materials from cotton lint.