Hyperspectral Reconstruction in Combination With a Crown Porcupine Optimization-Optimized Support Vector Regression (CPO-SVR) Machine Learning Model for Predicting the Total Acid Content of Daqu

The total acid content (TAC) of Daqu during fermentation is an important index for evaluating the quality of Daqu. In order to overcome the problem of low detection accuracy of RGB image and strict environmental requirements of HSI detection. Therefore, this study proposes a real-time and rapid detection method for the total acid content of Daqu by integrating spectral reconstruction technology with an optimized support vector regression (SVR) model. In this approach, RGB image data are acquired using an industrial camera, and hyperspectral data of the sample are generated via the MST++ reconstruction algorithm. These data serve as the input for the Daqu total acid content detection model. Additionally, the Crown Porcupine Optimization (CPO) algorithm is employed to optimize the parameters of the SVR model, thereby establishing a predictive model for the total acid content of Daqu. The experimental results show that the $R_p^2$ of the CPO-SVR model based on the reconstructed hyperspectral was 0.9449, the RPD was 4.2592, and the RMSEP was 0.0332. When compared to the CPO-SVR model based on original hyperspectral, the $R_p^2$ and the RPD were only 0.0185 and 1.0335 lower, while the RMSEP increased by 0.0062. The study showed that the MST++ hyperspectral reconstruction algorithm combined with the CPO-SVR model can realize real-time and rapid detection of the TAC of Daqu.