Journal of Near Infrared Spectroscopy

It was shown that linear transformations are suitable for use in multivariate calibration in near infrared spectroscopy as data compression tools. Partial Least Squares calibration models were built using spectral data transformed by expansion in the series of classical orthogonal polynomials, Fourier and wavelet harmonics. These models allowed effective prediction of the cetane number of diesel fuels, Brix and pol parameters of syrup in sugar production and fat and total protein content in milk. Depending on the compression ratio, prediction errors were no larger than 30% of corresponding errors obtained by the use of the non-transformed models. Although selection of the most suitable transformation depends on the calibration data and on the cross-validation method, in many cases Fourier transform gave satisfactory results.

Raw milk is often described as a major source of lactic acid bacteria for indigenous lactic starter. These indigenous starters contribute to the sensorial quality of cheese. Raw milk, rich in Lactoccocus lactis may therefore be very interesting for the cheese making. Currently, the most commonly used methods to differentiate lactic acid bacteria, and particularly the closely related phenotypes Lactococcus and Enterococcus, are based on DNA sequencing, but the cost and time required for these analytical methods hinder their use for rapid screening of raw material. The present study therefore proposes a simple alternative method to identify and discriminate against Lactococcus and Enterococcus, at the genus, but also at the species level, that is based on collecting near infrared spectra directly from bacterial colonies in Petri dishes. The infrared spectra of 280 strains of Lactococcus and Enterococcus cultured on solid media were collected by using a spectrometer with a wavelength range of 908 to 1684 nm and a remote probe. The best Classification And Regression Trees models for genus and species discrimination gave an excellent classification rate of 87% on an external validation set (30 strains). Loading line plots, with prominent bands at 900–960 and 1270–1390 nm, confirmed that the source of variation was due to changes in the polysaccharides.

Loblolly (Pinus taeda) and slash (Pinus elliottii) pine tree biomass are significant renewable energy resources for bioenergy industries in the Southeastern United States. There is a great need for evaluation of their properties (relevant to production and evaluation-screening for bioenergy extraction) by rapid but accurate analytical methods like near infrared reflectance (NIR) spectroscopy. However, no attempts have been made so far to develop useful NIR spectroscopic calibration models for analyzing biomass of these two species. In this study, acceptable NIR spectroscopic calibration models were developed for: gross calorific value, GCV (n = 181; R²= 0.83; ratio of standard error of cross-validation to deviation, RSCD = 2.32; ratio of standard error of cross-validation to inter-quartile distance, RSCIQ = 2.68); net calorific value, NCV (n = 184; R²= 0.83; RSCD = 2.31; RSCIQ = 2.69); ash-free calorific value, AFCV (n = 184; R²= 0.83; RSCD = 2.24; RSCIQ = 2.49); moisture (n = 181; R²= 0.85; RSCD = 2.54; RSCIQ = 2.30); ash (n = 180; R²= 0.86; RSCD = 2.65; RSCIQ = 2.62); total carbon, C (n = 101; R²= 0.95; RSCD = 3.01; RSCIQ = 5.66); total nitrogen, N (n = 87; R²= 0.95; RSCD = 4.24; RSCIQ = 4.76); and total sulfur, S (n = 83; R²= 0.97; RSCD = 4.17; RSCIQ = 3.05) contents of these biomasses though the calibration models for NCV and AFCV are indirect calibration. Prediction of the independent validation sets yielded good agreement between the NIR spectroscopic predicted values and the laboratory reference values for each of: GCV (n = 92; r²= 0.89; ratio of performance to deviation; RPD = 3.01; ratio of performance to inter-quartile distance, RPIQ = 3.16); NCV (n = 91; r²= 0.83; RPD = 2.43; RPIQ = 3.06); AFCV (n = 91; r²= 0.80; RPD = 2.25; RPIQ = 2.83); moisture (n = 92; r²= 0.82; RPD = 2.38; RPIQ = 2.40); ash (n = 89; r²= 0.81; RPD = 2.30; RPIQ = 2.66); C (n = 43; r²= 0.90; RPD = 3.14; RPIQ = 3.23); N (n = 44; r²= 0.95; RPD = 4.33; RPIQ = 5.96); and S (n = 42; r²= 0.93; RPD = 3.67; RPIQ = 3.24) contents, indicating that all eight calibration models had good quantitative information. The standard errors of prediction for all models were less than twice the corresponding standard error of laboratory. Therefore, precise, accurate, and rapid analysis of calorific values and C, N, S contents of these biomasses can be done using these novel NIR spectroscopic calibration models developed.

Pterocarpus erinaceus (Poir.) (local name: Toti, Tem) is a species endemic to the Guinea-Sudan and Sudan-Sahel zones. It is over-exploited for its wood, which is sold in international trade to China and Indian markets. The selection of high performance wood based on wood properties and their vegetative reproduction represents the best option for the establishment of plantations with the aim of promoting the restoration of degraded natural resources. However, measurements of these characters are lengthy, time consuming and destructive, and difficult to perform in Togo because of the lack of equipment. An alternative to those methods is near infrared (NIR) spectroscopy. The aim of this study was to assess the efficiency and feasibility of near infrared spectroscopy for predicting the physical and mechanical properties and the colour parameters of P. erinaceus wood from Burkina Faso, Niger and Togo. The properties were determined by conventional method and correlated with near infrared spectra using partial least square regression. The partial least square regression models were tested by cross-validation and by external validation. The results of this study showed the possible use of near infrared spectroscopy to estimate the physical and mechanical properties and the colour parameters of P. erinaceus wood. In fact, when validating prediction models for radial and tangential shrinkage, the fibre saturation point (FSP) showed R² values of 0.70, 0.71, and 0.92, respectively, with ratio performance deviation values of 1.84, 1.87 and 3.57. For the colour parameters L*, a* and b*, the R² values were 0.83, 0.74 and 0.67, respectively, with ratio performance deviation values of 2.40, 2 and 1.70. The results showed that the basic density and fibre saturation point models were classed as good to very good, respectively. Near infrared spectroscopy can therefore be used to predict efficiently the properties of P. erinaceus wood for a large number of samples, enabling the incorporation of those traits in breeding criteria.

Comparisons are reported for developing predictive models for dry matter across a wide variety of fruits with near infrared spectroscopy instrumentation, using a number of commercially available hand-held portable instruments (NIRVANA by Integrated Spectronics, F-750 by Felix Instruments, H-100C by Sunforest and SCiO by Consumer Physics) and an in-house laboratory based instrument (Benchtop). Three intrinsic (same fruit type) and combined (all fruit types) data sets were created from two separate batches of fruit populations. The first batch (Lot I) consisted of 205 ripe fruits from three different main fruit types (apples, kiwifruit and summerfruit) and 12 distinct fruit sub-categories. The second batch (Lot II) consisted of 91 ripe fruits from two different fruit types (apples and kiwifruit) and seven distinct fruit sub-categories. The laboratory based Benchtop instrument performed the best overall with typically higher prediction r² values (>0.92). The hand-held instruments delivered moderate to high r² values between 0.8 and 0.95. Results obtained with the intrinsic data sets revealed typically lower root mean square errors of prediction for apples and kiwifruit (0.32% to 0.73%) and larger prediction errors for summerfruit (0.53% to 0.82%). Some large performance variations between instruments of the same type were observed suggesting caution in evaluating the relative performance of different instrument types or formats on the basis of data generated with just a single instrument and/or data set. However, performance differences between the different hand-held portable instruments, on the same data sets, were often not statistically significant (p < 0.05). Instrument choice for any particular application will likely come down to matters not considered here, such as, for example, ease and accuracy during in-field operation and overall reliability.

Nine neural models were created to predict the characteristics of the extra virgin olive oil developed as a quality objective and by-products. These models are designed with the help of data of process variables from physical sensors such as temperature, flows, current intensity, etc. and physicochemical ones like the near infrared spectrum of the olive mass. The results obtained for the extractability of the process (fatty content and moisture) were highly significant correlations (r²≥0.90) and with similar prediction errors (root mean of squared error of prediction) relative to other analysis techniques which measure the by-product directly. For prediction the models gave correlations above 0.94, with the exception of ultraviolet absorption coefficients (0.72–0.84), with small prediction errors and the quality indicator relative error range with values above the optimal 10. The set of developed artificial neural networks models constitute the basis of the global ‘simulator’ tool of the extra virgin olive oil process. This simulator can perform a predictive optimization of the process to pre-adjust the process variables according to the goals marked in productivity or quality, from an near infrared spectral database or by real-time scanning. This simulator could be integrated into a control system that performs the function of a ‘virtual plant’ that allows the said system to adjust in real time the appropriate variables to meet the objectives.