Abstract: This study proposes a rapid detection method for Aspergillus ochraceus contamination in rice, combining near-infrared spectroscopy (NIRS) with machine learning algorithms, aiming to efficiently predict the total fungal count in stored rice. In this study, rice inoculated with A. ochraceus was used to simulate fungal infection during storage. Diffuse reflectance spectral data of rice samples at different stages of fungal growth were collected, and the fungal count was determined using the plate counting method. To improve the quality of the spectral data, moving average (MA), Savitzky-Golay (SG), standard normal variate (SNV), normalization (Nor), and the first derivative (FD) were applied for preprocessing to reduce noise and scattering effects. Additionally, three feature extraction methods, competitive adaptive reweighted sampling (CARS), successive projection algorithm (SPA), and genetic algorithm (GA), were used to select key wavelengths for further data optimization. Subsequently, partial least squares regression (PLSR), convolutional neural network (CNN), and extreme learning machine (ELM) were used for quantitative modeling of fungal counts, and their performance was compared. The results showed that MA and SG preprocessing effectively improved the signal-to-noise ratio of the spectra, thereby enhancing the predictive ability of the models. All three feature extraction algorithms successfully reduced the spectral dimensionality and identified key wavelengths related to the total fungal count. ELM showed significant advantages in handling nonlinear and high-dimensional spectral data, and the SG-GA-ELM model exhibited the best predictive performance (RC2 = 0.985 8, RP2 = 0.977 0, RMSEC = 0.167 2, RMSEP = 0.203 1), enabling high-precision prediction of the total A. ochraceus count. This study validates the feasibility of combining near-infrared spectroscopy with machine learning algorithms for the detection of A. ochraceus contamination in rice, providing a new, non-destructive, rapid, and intelligent monitoring approach for fungal contamination in stored grains. The proposed method holds great potential for applications in food safety monitoring, storage management, and online quality control of grains.

Key words: Aspergillus ochraceus; near-infrared spectroscopy; machine learning algorithms; non-destructive detection; quantitative prediction models