Nondestructive detection for Panax notoginseng powder grades based on hyperspectral imaging technology combined with CARS‐PCA and MPA‐LSSVM

Abstract This study investigated the feasibility of using hyperspectral imaging (HSI) technology to detect Panax notoginseng powder grades. The hyperspectral images of 240 Panax notoginseng powder samples were collected in the spectral range of 450–1,000 nm. Savitzky–golay (SG) and multiplicative scatter correction (MSC) were used to preprocess the original spectra. A method of combing competitive adaptive reweighted sampling (CARS) and principal component analysis (PCA) was used to analyze the spectral data and eliminate the influence caused by the randomness of Monte Carlo (MC) sampling. The least‐squares support vector machine (LSSVM) modeling results showed that CARS‐PCA had better spectral information extraction performance than PCA and CARS, and two principal components were extracted for modeling. The average classification accuracy of PCA‐LSSVM, CARS‐LSSVM, and PCA‐CARS‐LSSVM was 88.33, 90.93, and 92.5%, respectively. To further improve the modeling accuracy, a marine predators algorithm least squares support vector machine (MPA‐LSSVM) model was proposed to identify the grades of Panax notoginseng powder. The result indicated that MPA‐LSSVM had higher modeling accuracy and stronger robustness than the other compared models, and the classification accuracy of the training set and test set was 96.67 and 95%, respectively. The results illustrated the potential of HSI technology as an effective tool in Panax notoginseng powder grades detection. Practical Application This study verified the feasibility of using HSI technology to detect the grades of Panax notoginseng powder. Two principal components were extracted, which were used for simplifying the LSSVM model based on full wavelengths. The simplified model achieved high classification accuracy (95%), which provides a basis for the design of a multispectral imaging system.