Tomas Isaksson

The Effect of Multiplicative Scatter Correction (MSC) and Linearity Improvement in NIR Spectroscopy

Near-infrared (NIR) reflectance spectra of five different food products were measured. The spectra were transformed by multiplicative scatter correction (MSC). Principal component regression (PCR) was performed, on both scatter-corrected and uncorrected spectra. Calibration and prediction were performed for four food constituents: protein, fat, water, and carbohydrates. All regressions gave lower prediction errors (7–68% improvement) by the use of MSC spectra than by the use of uncorrected absorbance spectra. One of these data sets was studied in more detail to clarify the effects of the MSC, by using PCR score, residual, and leverage plots. The improvement by using nonlinear regression methods is indicated.

Related versions of the multiplicative scatter correction method for preprocessing spectroscopic data

Abstract Various multiplicative transformations of spectral variables have been used with some success as preprocessing methods for diffuse near-infrared spectroscopy data. We discuss first which additive/multiplicative transformations conserve the area under the spectral curve. Next we look at the relationship between the Multiplicative Scatter and the Standard Normal Variate transformations, and use a minimizing norm formulation to isolate three specific transformations that seem to be singled out as particularly interesting. In an empirical investigation using 8 different data sets, each with several constituents, the different transformation methods are compared.

On-line NIR analysis of fat, water and protein in industrial scale ground meat batches

Fat, water and protein contents in industrial scale meat batches were determined on-line by near infrared (NIR) reflectance spectroscopy. The NIR instrument was mounted at the outlet of a large meat grinder, and the measurements were performed in an industrial environment. Beef and pork samples, with chemical compositions of 7-26% fat, 58-75% water and 15-21% protein, were processed with hole diameters of 13mm in the grinder plate. Calibrations were made both for a combined set of beef and pork samples, and for separate sets of beef and pork samples. Validations were either done by full cross validation of the calibration set, or by bias corrected prediction of a test set. Prediction errors for the two sample sets, expressed as root mean square errors of cross validation or standard error of prediction, were in the ranges 0.82-1.49% fat, 0.94-1.33% water and 0.35-0.70% protein, depending of sample set and species of animal. The presented application is an improvement to the existing manual meat standardisation procedure, and has been implemented for regular use in a Norwegian meat manufacturing plant.

Piece-Wise Multiplicative Scatter Correction Applied to Near-Infrared Diffuse Transmittance Data from Meat Products

This paper presents a nonlinear scatter correction method, called piece-wise multiplicative scatter correction (PMSC), that is a further development of the multiplicative scatter correction (MSC) method. Near-infrared diffuse transmittance (NIT) data from meat and meat product samples were used to test the predictive performances of the PMSC and the MSC methods. With the use of PMSC, the prediction errors, expressed as the root mean square error of prediction (RMSEP), were improved by up to 36% for protein, up to 55% for fat, and up to 37% for water, in comparison to uncorrected data. The corresponding improvements by using PMSC compared to MSC were up to 22%, 24%, and 31% for protein, fat, and water, respectively.

Artificial neural networks in multivariate calibration

This paper is about the use of artificial neural networks for multivariate calibration. We discuss network architecture and estimation as well as the relationship between neural networks and related linear and non-linear techniques. A feed-forward network is tested on two applications of near infrared spectroscopy, both of which have been treated previously and which have indicated non-linear features. In both cases, the network gives more precise prediction results than the linear calibration method of PCR.

Locally Weighted Regression in Diffuse Near-Infrared Transmittance Spectroscopy

This paper presents an application of locally weighted regression (LWR) in diffuse near-infrared transmittance spectroscopy. The data are from beef and pork samples. The LWR method is based on the idea that a nonlinearity can be approximated by local linear equations. Different weight functions (for the samples) as well as different distance measures for “closeness” are tested. The LWR is compared to principal component regression and partial least-squares regression. The LWR with weighted principal components is shown to give the best results. The improvements with respect to linear regression are up to 15% of the prediction errors.

Differentiation of Frozen and Unfrozen Beef Using Near-Infrared Spectroscopy

Freezing and thawing affects the quality of meat. The present paper focuses on using near-infrared (NIR) diffuse reflectance spectroscopy to detect whether beef has been frozen and thawed. Intact beef and drip or centrifuged meat juice of M longissimus dorsi slices from 40 cattle were used as samples. The meat juices were analysed using dry extract spectroscopy by infrared reflection (DESIR). From centrifuged juice 80 samples were classified 100% correctly, using crossvalidation, into frozen or unfrozen beef by the K nearest neighbours method. This was obtained by high-order principal components from 400–2500 nm spectra. Other multivariate techniques, smaller wavelength ranges and detecting refrozen, thawed beef also gave results between 90 and 100%. Analyses of drip loss, exudative properties, water-holding capacity and dry matter of meat juice supported the interpretation of the NIR measurements. The results showed that NIR might be used as a screening method to differentiate unfrozen and frozen beef.

Quantitative determination of saturated-, monounsaturated- and polyunsaturated fatty acids in pork adipose tissue with non-destructive Raman spectroscopy.

The composition of dietary fat has received increased attention during the recent years because it influences human health. Seventy seven samples from pork adipose tissue and melted fat from the same tissue were measured with Raman spectroscopy. Gas chromatography analysis was conducted as reference. Iodine values (IV) ranged from 58.2 to 90.4g iodine added per 100g fat. Polyunsaturated fatty acids (PUFA) ranged from 7.8% to 31.7% and monounsaturated fatty acids (MUFA) from 35.2% to 51.5% of total fatty acids. When applied on pre-processed spectra of melted fat, partial least square regression (PLSR) with cross-validation gave a correlation coefficient (R)=0.98, and root mean square error of cross-validation (RMSECV)=1.4 for IV, using 3 PLS factors in the model. PUFA gave R=0.98 and RMSECV=1.0% of total fatty acids, using 5 PLS factors. MUFA were predicted with R=0.96 and RMSECV=1.0% of total fatty acids, using 9 PLS factors. On adipose tissue a model with 3 PLS factors gave R=0.97 and RMSECV=1.8 for IV. For PUFA, a model with 3 PLS factors gave R=0.95 and RMSECV=1.5% of total fatty acids. For MUFA a model with 6 PLS factors gave R=0.91 and RMSECV=1.5% of total fatty acids. The results indicate the feasibility to use Raman spectroscopy as a rapid and non-destructive method to determine IV, PUFA, MUFA and saturated fatty acids (SFA) measured directly on pork adipose tissue and in melted fat from the same tissue.

Near infrared reflectance spectroscopy in the prediction of sensory properties of beef

Near infrared (NIR) spectroscopy in the prediction of sensory hardness, tenderness and juiciness of bovine M. Longissimus dorsi muscles has been studied. Principal component regressions (PCR) of sensory variables from NIR reflectance measurements on frozen/thawed beef of 120 heat treated samples yielded multivariate correlation coefficients of cross-validation of 0.74, 0.70 and 0.61 for hardness, tenderness and juiciness, respectively. The corresponding correlation coefficients for NIR measurements of fresh (non-frozen) samples were approximately 0.1 units lower for all sensory variables. Predicting Warner Bratzler (WB) shear press values from NIR measurements gave a correlation coefficient similar to that for prediction of sensory hardness. The univariate correlation coefficient between sensory hardness and WB shear press values was 0.90.

Selection of Samples for Calibration in Near-Infrared Spectroscopy. Part II: Selection Based on Spectral Measurements:

Two strategies for selection of samples based on spectral measurements on a large set of samples are tested and compared. A method based on cluster analysis appears to be the best. The same prediction results achieved with the whole calibration set of 114 samples were obtained with only 20 samples selected by this algorithm.