Hyperspectral imaging combined with optimized SPA-GA and CPO-SVR machine learning models for the rapid determination of the total nitrogen content in cellar soil

Abstract

The fermentation of Baijiu grains in the cellar is significantly influenced by the quality of the cellar soil, which contains a diverse range of microorganisms and physicochemical components. Among these, the total nitrogen content (TNC) is a critical indicator of soil quality and thus requires real-time monitoring to ensure quality control of the Baijiu. In this study, we developed two optimized machine learning algorithms—successive projection algorithm-genetic algorithm (SPA-GA) and crown porcupine optimization (CPO) achieve the rapid and accurate detection of the TNC in cellar soil using hyperspectral imaging (HSI). The feature wavelengths were selected by combining the SPA with the GA. Subsequently, the support vector machine regression (SVR) algorithm was further optimized using the CPO algorithm to establish a prediction model for determining the TNC. Comparative analysis of the various models demonstrated that the CPO-SVR model based on the feature wavelength spectral data extracted by the SPA-GA exhibited the best performance (${R_p}^2$= 0.9958, root-mean square error of prediction [RMSEP] = 0.0073 g/100 g). This model reduced the number of wavelengths by 86.16%, increased the ${R_p}^2$ by 0.3014, and decreased the RMSEP by 0.0566 compared to the same model built using the full-wavelength spectral data. These results indicated that the GA significantly enhanced the feature extraction capability of the SPA, thereby improving the model accuracy while reducing the number of wavelengths to reduce computational load. Furthermore, CPO was introduced to optimize the SVR, yielding the optimal parameter combination, which further improved the prediction model performance and accuracy while mitigating artificial parameter-seeking instability. HSI, in conjunction with the optimization algorithms, offers a novel method for the rapid, non-destructive detection of total nitrogen and other components in cellar mud.