Petter Vejle Andersen

The aim of the present study was to critically evaluate the potential of using NIR and Raman spectroscopy for prediction of fatty acid features and single fatty acids in salmon muscle. The study was based on 618 homogenized salmon muscle samples acquired from Atlantic salmon representing a one year-class nucleus, fed the same high fish oil feed. NIR and Raman spectra were used to make regression models for fatty acid features and single fatty acids measured by gas chromatography. The predictive performance of both NIR and Raman was good for most fatty acids, with R² above 0.6. Overall, Raman performed marginally better than NIR, and since the Raman models generally required fewer components than respective NIR models to reach high and optimal performance, Raman is likely more robust for measuring fatty acids compared to NIR. The fatty acids of the salmon samples co-varied to a large extent, a feature that was exacerbated by the overlapping peaks in NIR and Raman spectra. Thus, the fatty acid related variation of the spectroscopic data of the present study can be explained by only a few independent principal components. For the Raman spectra, this variation was dominated by functional groups originating from long-chain polyunsaturated FAs like EPA and DHA. By exploring the independent EPA and DHA Raman models, spectral signatures similar to the respective pure fatty acids could be seen. This proves the potential of Raman spectroscopy for single fatty acid prediction in muscle tissue.

Raman spectroscopy is a viable tool within process analytical technologies due to recent technological advances. In this article, we evaluate the feasibility of Raman spectroscopy for in-line applications in the food industry by estimating the concentration of the fatty acids EPA + DHA in ground salmon samples (n = 63) and residual bone concentration in samples of mechanically recovered ground chicken (n = 66). The samples were measured under industry like conditions: They moved on a conveyor belt through a dark cabinet where they were scanned with a wide area illumination standoff Raman probe. Such a setup should be able to handle relevant industrial conveyor belt speeds, and it was studied how different speeds (i.e., exposure times) influenced the signal-to-noise ratio (SNR) of the Raman spectra as well as the corresponding model performance. For all samples we applied speeds that resulted in 1 s, 2 s, 4 s, and 10 s exposure times. Samples were scanned in both heterogenous and homogenous state. The slowest speed (10 s exposure) yielded prediction errors (RMSECV) of 0.41%EPA + DHA and 0.59% ash for the salmon and chicken data sets, respectively. The more in-line relevant exposure time of 1 s resulted in increased RMSECV values, 0.84% EPA + DHA and 0.84% ash, respectively. The increase in prediction error correlated closely with the decrease in SNR. Further improvements of model performance were possible through different noise reduction strategies. Model performance for homogenous and heterogenous samples was similar, suggesting that the presented Raman scanning approach has the potential to work well also on intact heterogenous foods. The estimation errors obtained at these high speeds are likely acceptable for industrial use, but successful strategies to increase SNR will be key for widespread in-line use in the food industry.

Raman spectroscopy (RS) has for decades been considered a promising tool for food analysis, but widespread adoption has been held back by, e.g., high instrument costs and sampling limitations regarding heterogeneous samples. The aim of the present study was to use wide area RS in conjunction with surface scanning to overcome the obstacle of heterogeneity. Four different food matrices were scanned (intact and homogenized pork and by-products from salmon and poultry processing) and the bulk chemical parameters such as fat and protein content were estimated using partial least squares regression (PLSR). The performance of PLSR models from RS was compared with near-infrared spectroscopy (NIRS). Good to excellent results were obtained with PLSR models from RS for estimation of fat content in all food matrices (coefficient of determination for cross-validation (R ² _CV ) from 0.73 to 0.96 and root mean square error of cross-validation (RMSECV) from 0.43% to 2.06%). Poor to very good PLSR models were obtained for estimation of protein content in salmon and poultry by-product using RS (R ² _CV from 0.56 to 0.92 and RMSECV from 0.85% to 0.94%). The performance of RS was similar to NIRS for all analyses. This work demonstrates the applicability of RS to analyze bulk composition in heterogeneous food matrices and paves way for future applications of RS in routine food analyses.

Dry matter (DM) content is one of the most important quality features of potatoes. It defines the physical properties of the potatoes and determines what kind of product the potatoes can be used for. This paper presents the results obtained by a novel prototype NIR (near-infrared) instrument designed to measure DM content in single potatoes in process. The instrument is based on interaction measurements to measure deeper into the potatoes. It measures rapidly, up to 50 measurements per second, allowing several moving potatoes to be measured per second. The instrument also enables several interactance distances to be recorded for each measurement. The instrument was calibrated based on three different potato varieties and the calibration measurements were done in a process plant, making the calibration model suitable for in-line use. A good calibration for DM was obtained by partial least squares regression (RMSECV = 0.78% DM, R² = 0.91). The instrument was tested in-line in the process plant and several batches of potatoes were monitored for the estimation of the DM distribution per batch. Accuracy of DM determination as function of measurement position on the potato was studied, and results indicate that NIR scans along the center part of the potatoes give slightly better results compared to scans taken on either side of the center. Small differences in optical measurement geometry influence the accuracy of the calibration models, underlining the importance of optimizing instrument design for successful measurements.