This study aims to develop a rapid and non-destructive method based on near-infrared (NIR) spectroscopy combined with machine learning modeling for the quantitative detection of soybean-adulterated coffee powder. A hierarchical modeling strategy was adopted to improve prediction accuracy. Support vector regression (SVR) combined with three spectral preprocessing methods was used to construct prediction models. A total of 30 characteristic wavelengths were selected by comparing competitive adaptive reweighted sampling (CARS) and iteratively retains informative variables (IRIV). Furthermore, three optimization algorithms: dung beetle optimization (DBO), particle swarm optimization (PSO), and grey wolf optimizer (GWO) were tested to find the most effective algorithm. The CARS-DBO-SVR model exhibited coefficients of determination (R²) of 0.9784 and 0.9669, root mean square error (RMSE) of 0.0157 and 0.0228, and residual prediction deviation (RPD) of 6.8096 and 5.4998 for the calibration and test sets, respectively. This study demonstrates that NIR spectroscopy provides an effective technical means for detecting soybean powder adulteration in coffee.

In order to identify Panax notoginseng powders from different root parts, an electronic nose and gas chromatography-mass spectrometry (GC-MS) were used to analyze the volatile components of the whole root powder, rhizome powder, taproot powder, lateral root powder and fibrous root powder of P. notoginseng. The data obtained were analyzed by multiple comparison. The statistical learning method was used to extract eight time-domain features from the response curves of the electronic nose, and correlation analysis was carried out. Three feature selection algorithms were used to reduce the dimension of the feature data. Classification models were built using support vector machine (SVM), least square support vector machine (LSSVM) or extreme learning machine (ELM) based on the original feature data or the three kinds of feature selection data. The grey wolf optimization (GWO) algorithm was introduced to optimize the parameters gam and sig2 in the classification model. The results showed that a total of 31 volatile compounds were detected in the five P. notoginseng powders. The best GWO-IRIV-LSSVM model could effectively distinguish the electronic nose data, with 97.5% accuracy for the test set. Moreover, the volatile composition of the five samples differed mainly in terms of the contents of total volatiles, alkanes, and aromatic compounds, which was consistent with the results of GC-MS. The method used in this study can be used for the detection of high-quality P. notoginseng powder from geo-authentic production areas mixed with low-quality P. notoginseng powder.

In order to rapidly and non-destructively detect gastrodin and 4-hydroxybenzyl alcohol in Rhizoma Gastrodiae, near infrared spectral data of the dried tuber of Gastrodia elata Bl. f. glauca S. Chow were collected in the wavelength range of 900-1700 nm. First, convolutional smoothing (SG) and standard normal variable transformation (SNV) were used for spectral data preprocessing. Second, feature wavelength extraction was carried out by competitive adaptive reweighted sampling (CARS) and iteratively retains informative variables (IRIV). According to the results of establishing least squares support vector machine (LSSVR) model based on feature wavelength, the best method of feature wavelength extraction was selected. In order to improve the accuracy of the model, this study introduced the artificial rabbits optimization (ARO) algorithm to optimize the regularization parameter γ and the kernel function density σ² in LSSVR and the superiority of ARO to particle swarm optimization (PSO) and grey wolf optimizer (GWO) was evaluated. The results showed that the ARO algorithm was superior to PSO and GWO in in terms of optimization speed and ability. The best prediction models for gastrodin and 4-hydroxybenzyl alcohol were CARS-ARO-LSSVR, with prediction correlation coefficient (R_p²) of 0.9696 and 0.9577, and root mean square error of prediction (RMSEP) of 0.014 and 0.020, respectively. Therefore, this study showsthat near-infrared spectroscopy can be used for quantitative detection of active components in Rhizoma Gastrodiae, which provides a theoretical basis for the development of rapid detection devices for Rhizoma Gastrodiae.