W. F. McClure

Fourier Analysis Enhances NIR Diffuse Reflectance Spectroscopy

Fourier coefficients computed from the NIR spectra of pulverized tobacco samples can be used to estimate certain chemical constituents in the samples. As few as 11 coefficients from the Fourier domain used in a stepwise multiple linear regression (SMLR) model provide results equivalent to a 7-term SMLR model using log 1/R from the wavelength domain. The Fourier model reduces the computation time for calibration by 96% compared to the wavelength model, and reduces the magnetic storage space requirements by 98%. Removing the mean term from the Fourier model partially corrects the particle size anomaly encountered in pulverized samples.

The application of fourier-transformed near-infrared spectra to quantitative analysis by comparison of similarity indices (CARNAC)

A new technique for utilising near-infrared spectroscopic data for quantitative analysis is proposed. The method uses a database of spectra stored in the form of the Fourier transform together with the associated chemical analysis. The spectrum of an unknown sample is compared to each member of the database and a small subset of very similar samples is isolated. The analyte value for the unknown is calculated from the analytical values of the subset. Results are given for the application of the method to the analysis of nicotine in tobacco.

Two-Dimensional Correlation of Fourier Transform Near-Infrared and Fourier Transform Raman Spectra I: Mixtures of Sugar and Protein

Two-dimensional (2D) correlation of near-infrared (NIR) and Raman spectra was carried out for mixtures of protein (lysozyme) and sugar (sucrose) to investigate the potential of this technique for qualitative NIR spectral interpretation. Cross-correlation by least-squares was employed to assess changes in both sets of spectra which result from changes in the set of sample spectra. Fourier transform (FT) NIR and NIR-excited FT-Raman spectra were measured for each of the samples under the same conditions, and point-for-point 2D cross-correlation was calculated. In this technique, each wavenumber in the NIR region gives rise to a sliced Raman spectrum where each data point is correlated to the NIR wavenumber, while each wavenumber in a Raman spectrum provides a sliced NIR spectrum in which each data point is correlated to the Raman wavenumber. For example, choosing NIR wavenumbers 7272, 6960, 6324, and 4812 cm−1 gives sliced Raman spectra with features attributable to sucrose, while choosing NIR wavenumbers at 8424, 5148, 5052, and 4584 cm−1 provides slices with distinct lysozyme features. Therefore, the technique permits the determination of the most probable origin of NIR signals by connecting NIR spectra, which have rather broad and overlapped bands, to Raman spectra consisting of sharp and clearly separated bands. It is also possible to produce sliced NIR spectra of lysozyme and sucrose by properly selecting wavenumbers in their Raman spectra. The NIR slices explain which wavenumbers in the NIR region are correlated to lysozyme or to sucrose. Thus, 2D correlation spectroscopy helps explain the reasons why certain wavenumbers are selected in a chemometric calibration model.

Hand-Held NIR Spectrometry: Part I: An Instrument Based upon Gap-Second Derivative Theory

A hand-held near-infrared (NIR) meter (called the Gmeter), based upon gap-second derivative (GSD) theory, was designed, constructed, and performance-tested. The design incorporated narrow-band interference filters for isolating the three wavelengths required by the GSD calculations. A microprocessor was included in the design to facilitate both stand-alone and personal computer (PC) operation. The Gmeter was mounted in a caddy for making measurements within the laboratory. Performance of the Gmeter was compared with the performance of a FOSS NIRSystems Model 6500 spectrophotometer for measuring protein in soy-protein/sugar mixtures and for measuring nitrogen in fescue grass tissue. Two calibrations were developed on both instruments: (1) single-term GSD equations and (2) three-term (log 1/R) multiple linear regression (MLR) equations. Second-derivative calibration experiments on the Model 6500 spectrophotometer formulated the basis for selecting the three filters in the Gmeter. Model 6500 data indicated that the GSD calibration [coefficient of variation (CV) = 5.14%] performed better than a three-term MLR equation (CV = 8.0%). In addition, the Gmeter performed almost as well (CV = 6.30%) as the Model 6500 (CV = 5.14%) for measuring protein in the mixtures using single-term GSD equations. An exciting extra in this study was the fact that measurements from the same three filters selected for determining protein in protein/sugar mixtures could be used for determining nitrogen (CV = 17.2%) in dry-grass tissue.

The Use of Trigonometric Polynomials to Approximate Visible and Near Infrared Spectra of Agricultural Products

A Fourier analysis of the spectra consisting of 1000 to 2000 reflectance values obtained by scanning agricultural products at specific, equally spaced wavelengths is used to reduce the amount of data that needs to be recorded. It is shown that as few as 50 pairs of coefficients retain the bulk of the information contained in the complete spectrum. Using the limited number of Fourier coefficients is equivalent to rejecting the high frequency noise without spectral distortion and loss of data points at the ends of the spectra.

Near infrared technology for precision environmental measurements: part 1. Determination of nitrogen in green- and dry-grass tissue

The driving force for this work is rooted in data that confirms the contamination of streams and lakes caused from excessive use of nitrogen, pesticides and other soil amendments. Traditional analytical (wet chemistry) methods are too slow and too costly for detecting ecological abuse. A technology that would characterise the nutritional status of growing plants in a timelier manner (preferably in real time as an applicator moves through the field) is needed to control the volume of amendments. This paper explores the potential of near infrared (NIR) spectrometry for measuring nitrogen in plant tissue. In particular, it discusses the development of nitrogen calibrations, and performance of those calibrations, for both green- and dry-grass tissue. Results, based on collaborative studies by several researchers indicate that nitrogen can be measured with an SEP of 0.411% and 0.167% for green- and dry-grass tissue, respectively.

Fourier self-deconvolution in the analysis of near infrared spectra of chemically complex samples

A demonstrated analytical tool for band enhancements in the mid-IR, Fourier self-deconvolution can serve similarly in the analysis of NIR spectra even when the spectra, obtained with dispersion instruments in the diffuse reflectance mode, have broad, overlapping bands not easily resolved. Unlike derivative enhancement methods, Fourier self-deconvolution preserves peak areas so that the deconvolted results may prove advantageous in using multilinear calibration techniques. Band enhancements of both synthetic and real spectra are discussed.

More on derivatives: resolving overlapping absorbance bands

Quantitative NIR calibrations are usually based upon spectra that have been pretreated in order to enhance calibration robustness. The gap second-derivative (GSD) is one of the more popular pretreatments. Textbooks and journal articles,1–3 demonstrate the effect of the GSD for resolving overlapping bands, correcting for baseline shift caused by particle size changes from one sample to another, eliminating the dominant linear trend characteristic of most NIR spectra, correcting for sample-to-detector position and other factors. A common approach to calibration, even when the spectra data are not very complicated, would be to allow the computer, using a stepwise multiple linear regression program (SMLR),4,5 to pick wavelengths from a set of 2nd derivative spectra. Trying this procedure on several data sets, it seemed strange to find the SMLR wavelength-selection algorithm shying away from negative absorbance peaks in favour of wavelengths on the slope of the bands or wavelengths on the positive lobe of the derivative spectra. What seems more surprising is that the seemingly off-peak artefacts often have very good correlation with chemistry and produce calibrations that have proven to be quite robust. This raises a valid question. Why does SMLR ignore the negative peaks, corresponding to absorption bands, in favour of points on the slopes of or on the derivative lobes? This article is a discussion of a study designed to answer this question. In an effort to simplify and clarify the mechanism involved, simple computer simulated spectra (see Figure 1) consisting of two absorbance bands were generated using CSAS.5 Figure 1(c) is a synthesised absorption spectrum resulting from the addition of two Gaussianshaped absorbance bands A and B. The resulting composite spectrum makes it difficult to stipulate the peak wavelength of band A. Yet, the two negative peaks in the GSD spectrum, despite peak shifting caused by the GSD pretreatment, clearly indicate the presence of two absorbers. Two things were considered in this study: (1) the distance between the two peaks of the absorbance bands and (2) the effects of the gap on the resulting spectrum.

Near infrared technology for precision environmental measurements: Part 2. Determination of carbon in green grass tissue

Composting is one of the most desirable techniques for reducing waste volume. To make good compost, the correct proportions of the elements carbon and nitrogen (30: 1 ratio) are important. In this paper, carbon quantification of green grass tissue using near infrared (NIR) technology was studied. Separate studies were conducted for the short-wavelength region (SWR = 700–1100 nm, a range that includes part of the visible spectrum) and long-wavelength region (LWR = 1100–2500 nm). Several spectral pretreatments (such as SNV, derivatives etc.) were implemented to optimise the stepwise multiple linear regression (SMLR) and partial least squares (PLS) calibrations. PLS analysis was conducted for all pretreatments. Results showed that the 2nd derivative of standard normal variate (SNV) pretreatment for the LWR and the SNV pretreatment for the SWR gave the best predictions. To simplify the PLS models, a weight index (WI), was defined as the absolute value of product between the regression vector from PLS analysis and the average spectrum. A simple PLS calibration was developed using selected peak wavelengths of regression vector with a minimum WI. The simple PLS models gave better results than the full PLS calibrations. According to this analysis, the C–H stretching of the first overtone at 1860 nm and the C–H stretching of the third overtone at 874 nm were the key bands for the SWR and LWR, respectively. SMLR analysis was performed on the same spectral data used in the PLS analysis. SMLR calibrations were developed using the key band chosen in PLS analysis. Although the performance of the calibrations were not as good as the PLS calibrations, the SMLR model produced acceptable calibrations for both the SWR and LWR. The simple fact that NIR technology can be used to determine both carbon and nitrogen very quickly makes it an ideal technology for monitoring material going into a composting operation.

NIR imaging spectroscopy: A fascinating frontier

Measurement of composition distribution is a fascinating idea. We have just completed the design and development of AGBOT (an AGricultural roBOT) specifically tailored for transferring cellgrown seedlings in a greenhouse environment. Greenhouse space is very expensive. Since germination rates may range from 70% to 90%, seedling transfers are necessary to ensure that a plant is growing in every cell. Robotics appears to be the answer to this tedious and timeconsuming process.