Non-Destructive Detection of Soybean Storage Quality Using Hyperspectral Imaging Technology

(1) Background: Soybean storage quality is crucial for subsequent processing and consumption, making it essential to explore an objective, rapid, and non-destructive technology for assessing its quality. (2) Methods: crude fatty acid value is an important indicator for evaluating the storage quality of soybeans. In this study, three types of soybeans were subjected to accelerated aging to analyze trends in crude fatty acid values. The study focused on acquiring raw spectral information using hyperspectral imaging technology, preprocessing by the derivative method (1ST, 2ND), multiplicative scatter correction (MSC), and standard normal variate (SNV). The feature variables were extracted by a variable iterative space shrinkage approach (VISSA), competitive adaptive reweighted sampling (CARS), and a successive projections algorithm (SPA). Partial least squares regression (PLSR), support vector machine (SVM), and extreme learning machine (ELM) models were developed to predict crude fatty acid values of soybeans. The optimal model was used to visualize the dynamic distribution of these values. (3) Results: the crude fatty acid values exhibited a positive correlation with storage time, functioning as a direct indicator of soybean quality. The 1ST-VISSA-SVM model was the optimal predictive model for crude fatty acid values, achieving a coefficient of determination (R²) of 0.9888 and a root mean square error (RMSE) of 0.1857 and enabling the visualization of related chemical information. (4) Conclusions: it has been confirmed that hyperspectral imaging technology possesses the capability for the non-destructive and rapid detection of soybean storage quality.