Fig 1.
The schematic of the system.
Fig 2.
The front panel of the custom software.
Fig 3.
The software architecture and the state diagram of the event-driven finite state machine.
Fig 4.
Cotton foreign matter and cotton lint samples used in this paper.
1-bark outer, 2-bark inner, 3-stem outer, 4-stem inner, 5, 6, and 7-brown leaf from DP 1050, NG5315, and PHY 339 respectively, 8, 9, and 10-bract from DP 1050, NG5315, and PHY 339 respectively, 11-hull, 12-twine, 13, 14, and 15-seed coat inner from DP 1050, NG5315, and PHY 339 respectively, 16, 17, and 18-seed coat outer from DP 1050, NG5315, and PHY 339 respectively, 19, 20, and 21-seed from DP 1050, NG5315, and PHY 339 respectively, 22-green leaf, 23-plastic bag, 24-plastic bale packaging, 25-paper, and 26, 27, and 28-cotton lint of DP 1050, NG5315, and PHY 339 respectively.
Fig 5.
The procedure of image acquisition, calibration, ROI selection and spectra extraction.
Table 1.
Calibration of the spectral accuracy using the standard spectra from Krypton, Xenon, and Hg(Ar) lamp.
Fig 6.
Single band images of eight brown trash at six representative wavelengths.
Fig 7.
Single band images of seven non-brown trash and lint at six representative wavelengths.
Fig 8.
The mean spectra (black solid line) and standard deviation (error bar) of eight brown trash.
Fig 9.
The mean spectra (black solid line) and standard deviation (error bar) of seven non-brown trash and lint.
Fig 10.
PCA score plot of 15 types of cotton trash and cotton lint.
(a) clusters under the top 3 PCs space, (b), (c), and (d) are projection on PC1 vs PC2, PC1 vs PC3, and PC2 vs PC3 respectively.
Fig 11.
p-values of the Hotelling paired-test using the feature set of top three PC scores.