Hlj August First Land Reclamation University

Sample heterogeneity compromises near-infrared (NIR) accuracy for crude fat content (CFC) in soybeans, traditionally requiring destructive grinding. Existing methods lack real-time nondestructive alternatives. We hypothesized that Raman-NIR fusion via an enhanced machine learning framework could enable precise, non-destructive CFC quantification. An enhanced broad learning system (EBLS) integrating NIR and Raman spectra was developed. The EBLS achieved accuracy comparable to destructive methods (RSQ > 0.92), outperformed existing grain-level techniques, and reduced moisture interference versus standalone NIR. Statistical validation confirmed no significant differences from national standards (95 % confidence). This multimodal spectral fusion paradigm allows real-time CFC analysis during soybean transactions and adaptive industrial parameter optimization, addressing preprocessing-free measurement needs.

The germination rate of maize seeds is a critical indicator for ensuring high-quality sowing and suitability for food processing. To address the limitations of traditional germination tests, a rapid and non-destructive evaluation method based on near-infrared (NIR) spectroscopy combined with Gaussian Process Regression (GPR) was developed. Various spectral data preprocessing techniques were applied, and a hybrid kernel function integrating Gaussian and Linear kernels was constructed. Particle Swarm Optimization (PSO) was used to optimize the kernel parameters. The PSO-GPR model achieved excellent performance, with determination coefficients (R²) of 1.000 and 0.9899 for the training and validation sets, respectively. The root mean square errors (RMSE) were 0.0059 and 0.0033, and the residual predictive deviation (RPD) reached 9.3, outperforming PLSR and SVM models. This study provides a novel strategy for the non-destructive evaluation of crop seed quality and contributes to developing smart agricultural practices.

Traditional conditional mutual information maximization (CMIM) algorithms struggled to capture nonlinear dependencies in continuous near-infrared (NIR) spectral analysis. Therefore, this study proposed a novel framework combining improved CMIM with SHAP for rapid maize crude fat content prediction. Innovatively replacing discretization with continuous variable estimates and two-stage filter-wrapper feature selection. The CMIM_KDE method achieved R²p of 0.7618 and 0.7531 in the PLSR and SVR models, respectively, which was an average improvement of 6.18 % and 4.42 % over other improved strategies. In addition, SHAP revealed the screened feature wavenumbers around 5684 cm^-1 and 4312 cm^-1 were related to the CH group of maize. The validity and generalizability of the method was validated on publicly available datasets. This study solved the applicability bottleneck of traditional CMIM method in NIR spectral analysis. It could provide technical support for the quality detection of other agricultural products as well as low-altitude remote sensing and real-time online prediction.