Low-Resolution Mid-Infrared Reflection Analysis for Discernment of Contaminants in Seed Cotton

Citation: Jiang W, Whitelock D, Hughs SE, Rayson G (2018) Low-Resolution Mid-Infrared Reflection Analysis for Discernment of Contaminants in Seed Cotton. Int J Analyt Bioanalyt Methods 1:001 Copyright: © 2018 Jiang W, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. *Corresponding author: Gary D Rayson, Department of Chemistry and Biochemistry, College of Arts and Sciences, New Mexico State University, Las Cruces, NM 88003-8001, USA, P.O. Box 30001, Tel: 575-646-5839, Fax: 575-646-2649


Introduction
Cotton is a very important natural fiber resource. Its production has an important impact on economies around the world. Higher cotton quality often increases its value and profitability [1]. However, contaminants mixed with cotton fiber during harvesting and processing dramatically decrease cotton's quality [2]. Cotton in the U.S. is widely harvested mechanically and packed using polymeric materials. Common contaminants in the cotton bale include both botanical components (e.g. leaf, stem) and synthetic materials (e.g. plastic bag, module cover) [3]. Identification and removal of contaminants in the cotton is a Abstract Contaminants mixed with cotton during harvesting and processing dramatically decreases its quality and economic value. A low resolution mid-infrared reflection instrument using four wavelengths (3100, 2900, 2300, and 1500 cm -1 ) was designed and constructed to distinguish cotton samples from 16 common contaminants (e.g. plastic and grease). These wavelengths were identified from associated high-resolution FT-IR spectra using multivariable analysis (i.e. Principal Component Analysis, Cluster Analysis, and Multiple Curve Regression). Simulation of low resolution spectra was undertaken to demonstrate feasibility of these wavelengths. Cotton and contaminants samples were analyzed in triplicate using the resulting in-house constructed instrument. Contaminants were successfully differentiated from cotton in a 33.2 mm 2 field of view with 100% accuracy. When mixed, limits of detection for positive identification of each contaminant was observed when the foreign material comprised between 11% and 60% of the field of view.
challenge. Although most large-sized cotton trash can be removed after ginning, this process can become more complicated due to the tendency of contaminants to be shredded to small fibers during processing [3,4].
Currently, trade and regulatory offices and laboratories in the cotton industry use both the High Volume Instrument (HVI TM ) and the Shirley Analyzer (SA) to measure cotton quality [5]. The HVI TM measures cotton length, strength, color, and trash content using visible imaging. It identifies the number of non-lint particles on a sample's surface and measures the surface area covered by nonlint particles [2,5]. The Shirley Analyzer (SA) is a | gravimetric technique to measure trash content using aero-mechanical methods to separate cotton fiber from trash. Although the HVI TM and the SA can measure some quality properties of cotton, they are limited to small cotton samples from each bale and lack specificity in the identification of different contaminant types in the cotton fiber [6,7].
Many researchers have tried different optical technologies to distinguish various types of contaminants, such as imaging approaches, ultraviolet-visible spectroscopy (UV-VIS) and Infrared spectroscopy (IR). Xu, et al. developed an imaging system based on a color CCD camera using color, shape and size attributes of cotton [7,8]. While a classification accuracy > 95% in classifying different trash types was reported, the results indicated the color feature was more reliable because contaminate shape and size varied during cotton processing. The limitation of this method was the prohibitive processing time, rendering it impractical for general cotton industry implementation [7]. Alternatively, Fortier, et al. investigated the capability of UV-VIS and Fouriertransform near-infrared reflectance (FT-NIR) for identifying botanical cotton trash types [3]. These yielded respective rates of botanical trash identification of 67% and 98%.
FT-NIR has also been widely used in cotton contaminants identification [3,4,6,8]. It is a rapid and accurate technique in identifying botanical cotton trash and synthetic materials. A NIR spectral library based on different botanical trash types has been generated enabling accurate cotton trash classification [5]. Fortier,et al. [6] also demonstrated an overall 98% accuracy in the botanical and synthetic material identification within pure cotton using a bench top FT-NIR instrument. Similarly, Himmelsbach, et al. successfully used  Table 1). algorithm was developed that enabled contaminate discernment using these low-resolution spectra. Limits of contaminate detection were then determined using this instrument.

Sample treatment and data analysis
Cotton samples (upland and pima) and common contaminants are shown in Figure 1 for clarification of each material analyzed. These are further described in Table 1 with reference sample numbers used throughout the study. Each of these samples was initially analyzed in triplicate using a commercial infrared spectrometer (Nicolet iS 10 FT-IR, Thermo Fisher, CA) with an attenuated total reflectance (ATR) sampling accessory. All spectra were collected using 1.0 cm -1 resolution and recorded as absorbance (i.e., -log(reflectance)) as a function of wavenumber from 400 to 4000 cm -1 . Representative spectra from a sample of cotton and each of five of the potential contaminates are shown in Figure  2. It should be noted that names used for each material are those commonly used within the cotton industry for clarity.
Principal component analysis (PCA) was applied to the resulting 54 spectra. This indicated 95.18% of the variance in the sample spectra could be described using five principal components. For illustration purposes, scores for the first three principal components (comprising 88.69% of the total variance) are shown in Figure 3a. Because five PCs were required to account for > 95% of the variance in the spectra, Cluster analysis (CA) was a FT-IR instrument with an Attenuated Total Reflectance (ATR) accessory to identify cotton botanical trash and synthetic materials [9]. The limitations of both FT-NIR and FT-IR have included the need of commercial instrumentation which can be prohibitively expensive. Samples must be collected and analyzed in the laboratory which is not conducive to their use in a processing facility [10,11].
The ability to accurately identify individual contaminants in cotton fiber at the ginning facility would greatly improve both the efficiency of contaminant removal and resulting fabric production quality [12]. Each of the techniques currently used to discern contaminates within seed cotton prior to ginning requires expensive instrumentation, off-site processing, or both. The goal of this work was to construct an inexpensive low resolution (only four specific wavelength regions) mid-infrared reflectance instrument to distinguish cotton fibers from synthetic materials. Ultimately, it could be applied to routine analysis within a ginning facility. In this work, FT-IR spectra were collected from a collection of common contaminates and those spectral regions enabling contaminate discernment were identified using chemometric tools. The feasibility of broad bandpass filters for each region was tested by generating simulated reflectance spectra using higher resolution FT-IR data. A reflection instrument was then design and constructed for collection of such low-resolution spectra for both upland and pima cotton samples with each of the same common contaminates. An Table 1: Samples analyzed by FT-IR in this study (The letters correspond to Figure 1. The numbers correspond to Figure 3a). undertaken to fully quantify separation of sample spectra projections into this 5-dimentional score space. The resulting dendrogram ( Figure 3b) clearly illustrates the separation of materials into different clusters. For example, cotton samples are clustered together while the different polymeric materials were placed in separate clusters. IR spectra of the pure cotton fibers could not be further distinguished (i.e. Pima and upland). However, the measured spectral signatures were able to clearly distinguish cotton fibers from any of the common contaminants ( Figure 1), including solid fiber and associated biological trash (e.g. plant stems), the objective of the present study. This demonstrated the ability to discern contaminate presence relative to cotton materials.

Number Sample names
Although PCA of FT-IR reflectance spectra separated cotton material from extraneous materials, each resulting Principal Component is a purely mathematical parameter describing the data variance and is without any physical significance [13] and did not reveal spectral regions responsible for their separation. It should be also noted that while the eigenvectors resulting from PCA do not exhibit the shape of the spectra of the components, features can be strongly related to each component. An alternative statistical tool enabling discernment of the contribution of each variable (i.e. wavelength) to variance within a sample set is Multivariate Curve Regression (MCR). MCR enables extraction of those spectral features enabling distinction among a collection of components. Thus, the derived spectrum of each factor responsible for sample clustering can be ascertained from the total data set [14]. To discern which portion of the FT-IR spectra was responsible for distinguishing contaminants from cotton, MCR was applied to the FT-IR reflectance data. Absorption bands at 3100, 2900, 2300, 1500, and 1000 cm -1 were identified as characteristic features describing each component. These bands correspond to a wing on the O-H stretching band (3100 cm -1 ), C-H stretching model (2900 cm -1 ), and C-C stretching (2300 cm -1 ) and bands at 1500 cm -1 and 1000 cm -1 indicative of vibrations within the fingerprint region of the spectrum (e.g., aliphatic C-H bending and carbohydrate C-O stretching). All spectra were compensated for atmospheric CO 2 near 2300 cm -1 using the ATR accessory. It was then hypothesized measurement of reflectance at these four or five wavenumbers could be sufficient for contaminant discernment.   Hz using an optical chopper and focused onto the sample using a 2.5 cm quartz plano-convex lenses (focal length 5.0 cm) with a second 4.0 cm plano-convex lens (focal length 7.0 cm. The sample holder (constructed in-house) was mounted on a x, y, z, θ stage. The translation stage was used for initial placement of the sample holder and remained constant for the duration of the study. Sample materials were pressed between a movable plate with a 0.65 cm aperture and a fixed base. The sample holder was blackened using candle soot to minimize background reflection detection. The reflected radiation was then imaged through each band-pass filter ( Table 2) using two additional plano-convex quartz lenses (2.54 cm diameter with focal lengths 3.81 and 5.0 cm, respectively) onto an IR detector with pre-amplifier (PVMI-2TE-1 * 1-TO8-BaF2, Boston Electronics Corporation, Boston). The detector output was processed with a lock-in amplifier (SR510 lock-in amplifier, Stanford Research, Stanford) referenced to the optical chopper. All optical components were determined to be transparent at each spectral band. specification for the Low-Resolution Mid-Infrared Reflection Spectrometer.

Data acquisition and analysis
Initial radiation intensity for each wavelength range I 0 was measured with no sample present and a neutral diffuse reflecting surface (Aluminum). Each sample was placed behind the aperture and the reflected intensity, I, measured. The reflectance, R was calculated by: R = I/I 0 And absorbance then calculated: Abs = -log(I/I 0 ) = -log(R) Each material was analyzed in triplicate and the average value recoded. Support information (S2) shows the averaged absorbance for each sample material and S3 shows the resulting low-resolution absorption spectra for each sample material. A mathematical model was then developed to separate contaminant materials from cotton samples.

Data Analysis Model
Given a set of data A = (X 1 , X 2 , · · · Xn) all coming from the cotton samples, where each data point is a 4-dimensional real vector, the method used these data to build a standard criterion for classifying the Simulation A simulation was then undertaken to distinguish among the 54 samples ( Figure 1 and Table 1) using four wavelength regions. (The 1000 cm -1 was excluded from further studies due to added complexity of light sources and detectors in this spectral region.) Commercially available IR filters (Multi IR optoelectronics, China) were identified for potential use in a low-resolution spectrometer.
To determine the feasibility of such bandpass filters for discernment of foreign materials within raw seed cotton, the transmission properties of these filters ( Table 2) were used for the generation of simulated low-resolution spectra using data from the higher resolution FT-IR spectrometer.
Transmission measurements using a highresolution FT-IR were then extracted for each filter bandwidth and applied to the corresponding filter transmission profile. The detailed calculation for simulation is in the supplement part (Supplementary Information). The resulting simulated reflectance spectrum for each material was then generated. PCA and subsequent CA were applied to the simulated spectra. The resulting score plot and corresponding cluster analysis dendrogram are shown in Figure 4a and Figure 4b, respectively. PCA results clearly show separation of cotton samples from other materials using three PCs from these low resolution simulated spectra. CA results also reveal simulated cotton samples as grouped in one cluster separated from other materials. The results indicate the potential for successful distinction between cotton and foreign materials using such low-resolution reflectance data. This indicated the feasibility for use of broad-band reflection measurements for discernment of contaminants. Figure 5 shows a schematic of the low-resolution IR reflection spectrometer. The incident infrared radiation (Infrared light source model 6580, Newport Corporation, MT) was modulated at 88  The program using this model was written in the Matlab (Mathworks, version 2013a, Natick, MA). Data from 30 cotton samples were used as standard samples. The spectra additional samples were then applied to this model. The results are shown in Table 3.

Instrument design and construction
Using this discernment model, only upland cotton stem and pima cotton stem samples were not successfully distinguished from cotton materials. This corresponds to a misclassification rate of only 11.1%. Because these were each pure material samples, a set of samples with varied amounts of contaminant within the sample window with a cotton background were generated and tested.
New module tarp sample was used to exemplify the calculated detectability. The model was used to determine if a sample was cotton or a contaminant. Image analysis software (ImageJ 1.49 v, National Institute of Health, US Department of Health and Human Services, Bethesda, MD) was used to calculate the area percentage of contaminant on a cotton sample. As indicated above, the aperture diameter in the sample holder plate was 6.5 mm (an area of 33.    (Figure 1) analyzed using the lowresolution mid-infrared reflection spectrometer.

Conclusion
In this study, a low resolution mid-infrared reflection instrument was designed and constructed for discernment of contaminants in seed cotton. This reflection instrument used only four IR filters with transmittance centered at 3100, 2900, 2300, and 1500 cm -1 to achieve the discernment of cotton samples from 16 contaminants (such as plastic and grease). These four spectral bands were determined from a high-resolution FT-IR spectra data base for all the samples (cotton and contaminants) using statistical tools (PCA, CA, and MCR). The instrument correctly differentiated contaminants from cotton with 100% accuracy as pure materials. When the contaminants were placed on a cotton background, new module tarp is reported as between 43.2%-31.2% (14.3-10.4 mm 2 ).

Detection limits
The detection limit of the instrument for each material was then calculated as the sampled area that resulted in a reflectance measurement that was three times larger than the standard deviation of the blank, i.e., S/N = 3. These limits of detection are listed in Table 4. Again, detectability was defined as the percentage of the field of view for the reflection measurement that contained the contaminant and resulted in a positive discernment of the contaminate. The remaining portion of the viewing area for each measurement consisted of pure cotton (Figure 6). Listed are the percentages of the viewing area containing the contaminant that resulted in positive discernment of its presence (as Table 4: The detect limit of each sample (Figure 1) using the Low-resolution mid-infrared reflection spectrometer (S/N = 3   the detection limit for all samples ranged from 60% to 11% of the field of view. The described instrument has significantly lower cost and time required for the task compared to use of a highresolution FT-IR spectrometer. More importantly, it has potential for application to gin facilities to separate contaminants from cotton in real time. This would dramatically improve its quality and textile utility of processed seed cotton.  The calculated total reflected radiation for the filter was calculated using