可见光谱代理和纹理特征结合优化回归预测中国枸杞中的关键矿物质

Chinese wolfberry (CW) is a medicinal-food homologous fruit. Key mineral elements are important indicators of the nutritional quality and geographical authenticity of CW. Rapid and non-destructive evaluation of key mineral elements remains challenging because inductively coupled plasma mass spectrometry (ICP-MS) is accurate but destructive and laboratory-dependent. In this study, 120 CW batches from Ningxia, Qinghai, Gansu, and Xinjiang were analyzed by combining ICP-MS reference measurements of Cu, Fe, Mn, and Zn with standardized RGB image features. Visible-color proxy curves were reconstructed from RGB/L*a*b* information and Gaussian fitting, and frequency-domain texture descriptors were extracted from Fourier-derived angle-energy curves. Multivariate analysis showed that mineral profiles, the 570-593 nm visible-color proxy band, and the 0-40° and 63-140° texture-angle ranges contributed to regional differentiation. A Dung Beetle Optimizer-radial basis function (DBO-RBF) regression model was then used for internal prediction of mineral element contents and compared with nine baseline regression algorithms. DBO-RBF achieved test-set R2 values of 0.949, 0.965, 0.951, and 0.913 for Cu, Fe, Mn, and Zn, respectively, with corresponding test RPD values of 2.9017, 3.2419, 3.0175, and 2.8416. These results indicate that image-derived visible-color and texture features can provide useful screening information for mineral quality assessment in CW, offering a rapid, low-cost, and non-destructive approach for preliminary batch evaluation.