Feasibility of compact near-infrared spectrophotometers and multivariate data analysis to assess roasted ground coffee traits

Jamille Carvalho Souza,Celio Pasquini,Maria C. Hespanhol

Food Control(2022)

引用 4|浏览2
Two low-cost, miniaturized near-infrared (NIR) spectrophotometers working in two different spectral ranges (900–1700 nm and 1300–2500 nm) and employing different spectrometric technologies were evaluated to assess several traits of ground roasted commercial coffees, using multivariate analysis. The comprehensive sample set comprised one hundred forty-five samples from different origins, blends of varietals (arabica/arabica, and arabica/robusta) or pure varietal (arabica), decaffeinated or regular, presenting several grades of sensorial traits (bitterness, intensity, acidity, body, and roast degree), as informed by the manufacturing company. Representative spectra of the samples were acquired using a probing device providing rotational and linear displacement of the sample vial. The spectral data set was evaluated using principal component analysis (PCA) models aiming to verify the discrimination capability of both instruments. When evaluated down to the fourth principal component by both compact instruments, origin, varietal, decaffeinated, and regular coffees are grouped and well-differentiated by the PCA scores. Differences among batches of the same coffee brand can be observed. Quantitative models based on partial least square (PLS) regression were evaluated to assess the feasibility of miniaturized spectrophotometers to determine the samples' intensity, bitterness, body, roast, and acidity traits. These sensorial traits can be predicted by partial least square models with root mean square errors of external validation of 1.3, 0.7, 0.6, 0.5, and 0.8 and 1.2, 0.7, 0.5, 0.5, and 0.7 for intensity, bitterness, body, roast, and acidity, respectively, using the data sets obtained by the NanoNIR, and the NeoSpectra instruments and employing variables selected by the Jack-Knife algorithm. Quantitative results can be improved using local regression models constructed with a restricted set of samples. Overall, the results reveal the potential of miniaturized, low-cost NIR spectrophotometers to characterize commercial ground roasted coffee qualitatively and quantitatively with a slight advantage of using the instrument providing the more comprehensive wavelength range and based on Fourier transform technology.
Near-infrared spectroscopy,Compact near-infrared spectrophotometers,Roast ground coffee,Coffee traits determination,Commercial coffee classification,Multivariate analysis
AI 理解论文