W hen attempting to move an assay from the laboratory, whether it is for research or quality assurance, into the processing plant environment or field, there are some vital questions that must be answered. First and foremost—is the instrument robust enough for the environment? Many laboratory-based instruments cannot operate in the harsh environment of the plant floor without hardening in some way. This greatly increases the already significant weight of the instrument and the cost of the installation. Second, can the instrument make the needed measurement? Many hand-held and portable instruments are limited in their capacities in one way or another. They may lack the required spectral range, resolution or some specific sampling geometry. Third, if the instrument goes to the field, it will require special scanning modes not normally found on benchtop units. In addition, weight and power consumption can quickly become critical issues. This paper discusses the issues involved in field and plant analyses and describes a new instrument which fills many of the needs for at-line, in-line and field applications. The capabilities, robustness, tested applications and sampling accessories will be described along with the similarity of duplicate systems to provide seamless calibration transfer from lab to line to field.

SUMMARY Eighteen grass samples (four tropical species harvested after 4 and 8 weeks regrowth, and six temperate species after 4 weeks regrowth sum- mer and fall cuttings) were analyzed for crude protein, ether extract, ash, neutral detergent fiber (NDF), acid detergent fiber (ADF), hemi- cellulose, holocellulose, and permanganate lig- nin (PML). The relationship between chemical composition and in vitro dry matter digestibil- ity (IVDMD) was investigated by regression methods. In vitro digestibility of isolated NDF, ADF and holocellulose was determined and compared to the digestibility of whole forage. The results indicated that differences in the cell wall matrix of tropical and temperate grasses could cause discrepancies in predicting digesti- bility from chemical compositional data. The best regression equation for predicting the in vitro digestibility of all the grass samples contained terms for protein, hemicellulose and PML. The ADF from tropical grasses was more digestible than the ADF isolated from temper- ate species. Lignin was important as a predictor of digestibility and appeared in six of nine equations, particularly in the equations for temperate grasses. Protein was the best predic- tor of digestibility for tropical (r = .90), but not for temperate species (r = -.17). One important compositional difference was the amount of hemicellulose, which was more abundant in the tropical than in the temperate grasses (tropical, 30 to 35% vs temperate, 22 to 27%).

The quality of flax fiber in the textile industry is closely related to the wax content remaining on the fiber after the cleaning process. Extraction by organic solvents, which is currently used for determining wax content, is very time consuming and produces chemical waste. In this study, near-infrared (NIR) spectroscopy was used as a rapid analytical technique to develop models for wax content associated with flax fiber. Calibration samples ( n = 11) were prepared by manually mixing dewaxed fiber and isolated wax to provide a range of wax content from 0 to 5%. A total of fourteen flax fiber samples obtained after a cleaning process were used for prediction. Principal component analysis demonstrated that one principal component is enough to separate the flax fibers by their wax content. The most highly correlated wavelengths were 2312, 2352, 1732, and 1766 nm, in order of significance. Partial least squares models were developed with various chemometric preprocessing approaches to obtain the best model performance. Two models, one using the entire region (1100–2498 nm) and the other using the selected wavelengths, were developed and the accuracies compared. For the model using the entire region, the correlation coefficient ( R 2 ) between actual and predicted values was 0.996 and the standard error of prediction (RMSEP) was 0.289%. For the selected-wavelengths model, the R 2 was 0.997 and RMSEP was 0.272%. The results suggested that NIR spectroscopy can be used to determine wax content in very clean flax fiber and that development of a low-cost device, using few wavelengths, should be possible.

The major question about the newer instruments is “are they more capable than the instruments we have been using for decades”? In looking for a way to examine the utility of commercial instrumentation capable of performing the measurement of stickiness in cotton, a method was required for evaluating the way these instruments performed compared with the research grade spectrometer which was first successfully used to measure stickiness. While all of the instruments had the potential analytical capability of measuring sticky cotton, all would require even more costly hardening to be used in a cotton gin or spinning plant. Regardless, the calibration would need to be transferred to any of these new instruments. While there are existing algorithms to do this, it is imperative to understand the differences between instrument platforms. Two-dimensional correlation spectroscopy (2DCOS) has been used to characterise instrumental differences which will affect performance, stability and reliability. The 2DCOS programs were re-written in Matlab ver. 6.51 to provide a common format with the newer instruments and the special purpose instruments which utilise Matlab for data functions. The instruments used in the complete calibration studies of sticky cotton were evaluated for their instrument function parameters. The possibilities for using 2DCOS for instrument characterisation was explored in concert with available spectral technologies.

In looking for a way to examine the utility of commercial instrumentation capable of performing the measurement of stickiness in cotton, a means of evaluating the way these nine instruments performed against the research grade spectrometer was required. While all of the instruments had the potential analytical capability of measuring sticky cotton, some were more expensive than the industry would afford and would require even more costly hardening to be used in a cotton gin or spinning plant and some were concept commercial instruments without published performance data. Regardless, the calibration would need to be transferred to any of these new instruments. While there are existing algorithms to do this, it is imperative to understand the differences between instrument platforms. Two-dimensional correlation spectroscopy (2DCOS) has been used to characterize instrumental differences which will affect performance, stability and reliability. The 2DCOS programs were re-written in MATLAB ver. 6.51 to get on a common format with the newer instruments and the special purpose instruments which utilize MATLAB for data functions. The instruments used in the complete calibration studies of sticky cotton and those that we examined as possible rugged small instruments were evaluated for their instrument function parameters. The possibilities for using 2DCOS for instrument characterization will be explored for the small instruments.

ABSTRACTThe classification of cereals using near‐infrared Fourier transform Raman (NIR‐FT/Raman) spectroscopy was accomplished. Cereal‐based food samples (n = 120) were utilized in the study. Ground samples were scanned in low‐iron NMR tubes with a 1064 nm (NIR) excitation laser using 500 mW of power. Raman scatter was collected using a Ge (LN2) detector over the Raman shift range of 202.45~3399.89 cm‐1. Samples were classified based on their primary nutritional components (total dietary fiber [TDF], fat, protein, and sugar) using principle component analysis (PCA) to extract the main information. Samples were classified according to high and low content of each component using the spectral variables. Both soft independent modeling of class analogy (SIMCA) and partial least squares (PLS) regression based classification were investigated to determine which technique was the most appropriate. PCA results suggested that the classification of a target component is subject to interference by other components in cereal. The Raman shifts that were most responsible for classification of each component were 1600~1630 cm‐1 for TDF, 1440 and 2853 cm‐1 for fat, 2910 and 1660 cm‐1 for protein, and 401 and 848 cm‐1 for sugar. The use of the selected spectral region (frequency region) for each component produced better results than the use of the entire region in both SIMCA and PLS‐based classifications. PLS‐based classification performed better than SIMCA for all four components, resulting in correct classification of samples 85~95% of the time. NIR‐FT/Raman spectroscopy represents a rapid and reliable method by which to classify cereal foods based on their nutritional components.

"Stickiness" in cotton is a major problem affecting throughput in cotton gins and spinning mills alike. Stickiness is thought to be caused by the deposition of sugars by insects, principally aphid and whitefly, on the open boll. Fourier transform near-infrared (FTNIR) spectroscopy was used to develop models for sugar content from high-pressure liquid chromatography (HPLC), thermodetector, and mini-card data. A total of 457 cotton samples were selected to represent both Upland and Pima varieties and cotton processing before and after ginning. The Unscrambler was used to develop the models. A successful model was made to determine the mini-card value and successfully detect "stickiness". The standard error of cross-validation (SECv) was 0.26 with an R-2 of 0.96. The model was not improved by increasing the range of "stickiness" as measured by the mini-card from the usual 0-3 scale to a scale of 0-8. If a value is determined to be greater than 1 it will be difficult to blend bales at a spinning plant "opening line" to allow for maximum efficiency of spinning.

Flax fibers may be blended with cotton to provide an aesthetic property, improve performance and tailor fabric properties. The quality and cost of the woven fabric blends are affected by the amount of linen in the blend. Microscopic and chemical analyses are currently used to determine linen content in fabrics. This study describes a method to predict the linen percentage in linen/cotton blends using Fourier transform near-infrared (FT-NIR) spectroscopy, rapidly and non-invasively. A calibration model using partial least squares regression analysis was developed with gravimetrically measured ground flax-cotton fiber mixtures as reference samples versus NIR spectra. The best model occurred with a combination of multiplicative scatter correction and first derivative processing of the spectral data gave a standard error of validation of 2.20%, and only one factor was used for the model performance. Using this model, flax content was predicted in specific mixtures of flax and cotton fibers, blended flax-cotton yarns, and various non-scoured flax-cotton fabrics, giving standard errors of prediction less than 3%. Application of the calibration model to the scoured fabric, however, resulted in a higher error value. This result seemed to be due to loss in wax components and substantial changes in the NIR absorbance values of the fabric resulting from the scouring process. An alternative calibration model for scoured and dyed fabrics was developed, and using the model it was possible to predict flax contents in dyed fabrics with an error of 4–6%.

Dietary fiber is an important quality parameter of barley (Hordeum vulgare L.) but is extremely laborious to measure. Near-infrared (NIR) transmission and reflectance spectroscopy were investigated as rapid screening tools to evaluate the total dietary fiber content of barley cultivars. The Foss Grainspec Rice Analyzer and NIR Systems 6500 spectrometer were used to obtain transmission and reflectance spectra, respectively, of polished grains and ground barley. Total dietary fiber was determined for each cultivar by AOAC Method 991.43. Modified PLS models developed for predicting total dietary fiber, using transmission spectra (850-1048 nm) of polished grains, had a standard error of cross validation (SECV) of 10.4 (range 58-197) g kg(-1) and R-2 of 0.82 indicating sufficient accuracy for selecting or rejecting high dietary fiber cultivars. NIR reflectance spectroscopy (1104-2494 nm) of ground barley samples resulted in a model with SECV of 5.2 (range 58-197) g kg(-1) and R-2 0.96, indicating a high degree of precision in the prediction of total dietary fiber. The increased accuracy of the reflectance model may be due in part to more information available in the wavelength region used. The precision, low cost per sample and speed of measurement of the technique allow making dietary fiber selection decisions for large numbers of progeny in barley breeding programs.

Amylose and amylopectin are two major carbohydrates in cereal and cereal food products. Both of these polysaccharides have complex conformations that affect their physical, chemical, and biological activities,1 despite the fact that they are mainly made up of a-(1→4)-linked Dglucose residues. Generally, amylose has been considered to be a linear polymer through a-D-(1→4) glycosidic linkages (Fig. 1), although now there is evidence that amylose is not completely linear. Amylopectin is a branched polymer, and a branch point occurs approximately every 20–25 glucose units when a chain of a-D(1→4) glucose units is linked to the C-6 hydroxymethyl position of a glucose molecule through an a-D-(1→6) glycosidic linkage. Thus, about 4–5% of the glucose units in amylopectin are involved in branch points. Clearly, both the relative proportions of amylose to amylopectin and a-D-(1→6) branch points depend on the source of the starch.2 Raman spectroscopy has found considerable applica-

Flax must be retted, in which bast fibres are separated from non-fibre components, and then mechanically processed to clean the fibres before industrial application. In the USDA Flax Fiber Pilot Plant, flax is first cleaned through four separate modules and then passed through a Shirley Analyzer to further clean fibres for high-value applications such as textiles. Often, multiple passages through the Shirley Analyzer are employed to obtain higher quality fibres, but it is difficult to determine when the limit for cleanliness is reached by this method. Further, it is clear that materials other than the woody shive components are being removed by Shirley-cleaning, and a method is needed to assess cleanliness beyond the measure for shives. In this study, we attempted to establish an index to determine the degree of purity of flax fibre during the secondary cleaning stage for high quality fibre. Dew-retted (DR) flax and enzyme-retted (ER) flax, which had been first processed through the USDA Flax Fiber Pilot Plant and assessed for shive content, were processed with 10 repetitions of cleaning through the Shirley Analyzer. For both flax samples, absorbances at 1730, 1766, 2312 and 2350 nm decreased with successive Shirley-cleaning steps. These wavelengths appeared to originate from the epidermal layer (EL) that was associated with the flax fibre, an index was calculated using 11 training samples and validated using 10 independent test samples from the same flax samples. Index values gradually decreased with successive Shirley-cleaning steps for both retted flax samples; a lower index value indicated cleaner fibre. Different curves were apparent for the two flax samples, suggesting variations in the cleanliness of the starting material or perhaps differencess in fibre composition. The results suggest it is possible to determine the extent of cleaning of flax fibre using NIR spectroscopy beyond that for shive content based on the epidermal layer of the plant.

The amount of energy derived from fat in foods is a requirement of U.S. nutrition labeling legislation and a significant factor in diet development by health professionals. Near-infrared (NIR) spectroscopy has been used to predict total utilizable energy in cereal foods for nutrition labeling purposes, and in the current study, was investigated as a method for evaluation of the amount of energy derived from fat. Using NIR reflectance spectra (1104-2494 nm) of ground cereal samples and reference values obtained by calorimetry and by calculation, modified PLS regression models were developed for the prediction of percent energy from fat and energy from fat/g. The models were able to predict the percent of utilizable energy derived from fat with SECV and R(2) of 1.86-1.89% of kcal (n = 51, range 0-43.0) and 0.98, respectively, and SEP and r(2) of 1.74% of kcal (n = 55, range 0-38.0) and 0.98, respectively, when used to predict independent validation samples. Results indicate that NIR spectroscopy provides useful methods for predicting the energy derived from fat in food products.

Shive is the main contaminant in flax fibre and affects fibre quality. In this study, we developed a calibration for determining shive content in flax using near infrared (NIR) spectroscopy and applied the model to pilot plant processed flax to predict shive content. The model based on "ground" mixtures performed best from multiplicative scatter correction after a second derivative treatment of the spectral data, giving a standard error of cross-validation of 0.35% using five factors. Prediction samples were Jordan enzyme(ER) and Natasja dew-retted (DR) flax that was collected after various stages of processing. When the model was applied to the "ground" flax, a high correlation was obtained between the NIR predicted value and actual shive content, giving a correlation coefficient of >0.98 for both retted flax samples. However, when the model was applied to the "as-is" flax, a slope and bias were observed. These deviations were corrected by a linear regression between predicted values of "ground" and "as-is" flax. For the NIR analysis of ER flax, the shive content decreased rapidly by the third processing step to 4 to 5% and almost 0% after the last step. For the DR flax, the shive content continuously decreased with processing to about 5% after the last step. The results indicate that it is possibile to measure shive in flax on a commercial processing line.

Generalized two-dimensional (2D) correlation analysis of visible/near-infrared (NIR) spectra was performed to characterize the spectral intensity variations of chicken muscles induced by either storage time/temperature regime or shear force values. The results showed that intensities of two visible bands at 445 and 560 nm increase with the storage temperature under identical treatment, possibly indicating a color change due to frozen storage. The 2D NIR correlation spectra indicated that all NIR bands reduce their spectral intensities, probably due to the water loss and compositional alterations during the freeze-thaw process as well as the tenderization development in muscle storage. The heterospectra correlating the spectral bands in both visible and NIR regions exhibited a strong correlation and suggested the sequential change between color and other developments in muscles. In addition, shear value-induced NIR spectral intensity variations detected significant differences in spectral features between tender and tough muscles.

Fourier-transform Raman (FT-Raman) spectroscopy and near-infrared (NIR) reflectance spectroscopy were used to compare calibration models for determining rice cooking quality parameters such as apparent amylose and protein. Samples from two seasons were used in each calibration set. The laboratory values ranged from 4.89 to 12.48% for protein and from 0.2 to 25.7% for amylose. The data for both FT-Raman and NIR were preprocessed with orthogonal signal correction (OSC) for standardization. For both spectroscopic methods, five models were optimized by partial least squares regression (PLSR) and by Martens' uncertainty regression (MUR), including no processing, smoothing, normalization, first derivative (D1), and second derivative (D2). Based solely on standard error of cross-validation (SECV), the FT-Raman method was superior to the NIR method for protein. For amylose, the FT-Raman and NIR methods resulted in similar calibration statistics with a high precision, with the, FT-Raman requiring fewer factors. The best FT-Raman models were generated from OSC preprocessing with MUR for protein (SECV 0.15%, five factors) and from OSC without MUR for amylose (SECV 0.70%, seven factors). The best NIR models were obtained with D2 transform of OSC spectra for protein (SECV 0.22%, four factors) and with OSC spectra for amylose (SECV 0.57%, 11 factors).

ABSTRACTThe use of the derivative method for near‐infrared (NIR) calibration was investigated to determine protein and amylose content in rice flour. Samples for two years, 1996 and 1999, were combined to give a wide range of the constituents for development of the calibration model. The NIR spectral data were transformed with Savitzky‐Golay derivative with multiplicative scatter correction. To develop the best derivative models, the polynomial fits (quadratic, cubic, and quartic), convolution intervals (3–11 points for protein, 3–17 points for amylose), and derivative orders (1st derivative D1; 2nd derivative D2) were investigated. For the protein analysis, all polynomial fits with 3–11 points were acceptable to develop both the D1 and D2 models. However, the three‐point quadratic and five‐point quartic fits were not acceptable for the D1 model, and the three‐point quadratic fit was not acceptable for D2. For the amylose analysis, the D1 model produced generally better results than D2. Higher convolution intervals were required for the D2 model, whereas the D1 model was not affected by convolution intervals. A quadratic (or cubic) fit with 17‐point convolution interval was acceptable for the amylose D2 model, and the quadratic fit with 5–11 points and cubic (or quartic) fit with 7–17 points were suitable for the D1 model. Based on the standard error of cross‐validation (SECV), the calibration models developed using data for two years resulted in good precision with an SECV of 0.23% for protein using four factors and an SECV of 1.0% for amylose using 10 factors.

This work attempted to interpret the principal component loadings spectra of principal component analysis on large spectral data sets with multi-variables using two-dimensional (2D) correlation analysis. Three examples of visible/near infrared (NIR) spectra of chicken muscles under different conditions were given and discussed. 2D analysis indicated that characteristic bands from loadings spectra are in good agreement with those from a small number of spectra induced by simple external perturbations. Although some advantages of 2D correlation analysis (such as sequential changes in intensity) were not available, it might still be useful for the understanding of large and complex spectral data sets with multi component variations.

Shive, the nonfiberous core portion of the stem, in flax fiber after retting is related to fiber quality. The objective of this study is to develop a standard calibration model for determining shive content in retted flax by using near-infrared reflectance spectroscopy. Calibration samples were prepared by manually mixing pure, ground shive and pure, ground fiber from flax retted by three different methods (water, dew, and enzyme retting) to provide a wide range of shive content from 0 to 100%. Partial least-squares (PLS) regression was used to generate a calibration model, and spectral data were processed using various pretreatments such as a multiplicative scatter correction (MSC), normalization, derivatives, and Martens' Uncertainty option to improve the calibration model. The calibration model developed with a single sample set resulted in a standard error of 1.8% with one factor. The best algorithm was produced from first-derivative processing of the spectral data. MSC was not effective processing for this model. However, a big bias was observed when independent sample sets were applied to this calibration model to predict shive content in flax fiber. The calibration model developed using a combination sample set showed a slightly higher standard error and number of factors compared to the model for a single sample set, but this model was sufficiently accurate to apply to each sample set. The best algorithm for the combination sample set was generated from second derivatives followed by MSC processing of spectral data and from Martens' Uncertainty option; it resulted in a standard error of 2.3% with 2 factors. The value of the digital second derivative centered at 1674 nm for these spectral data was highly correlated to shive content of flax and could form the basis for a simple, low-cost sensor for the shive or fiber content in retted flax.