Assessment of near-infrared hyperspectral imaging for the discrimination of naturally aflatoxin-contaminated nuts.

Aflatoxins (AFs) and ochratoxin A (OTA) are major safety concerns in nuts due to their toxicity and uneven distribution. To address this, near-infrared hyperspectral imaging (NIR-HSI) was evaluated as a rapid, non-destructive tool for single-kernel classification, complemented by a preliminary HPLC-FLD screening. In this study, 52 peanut and 33 pistachio samples of commercial origin were analyzed for AFs and OTA using an HPLC-FLD method. The occurrence of AFB1 detected in peanut samples was 7.70%, whereas OTA was detected in pistachio samples with an occurrence of 3.03%. The maximum concentrations observed were 0.60 μg/kg for AFB1 and 0.68 μg/kg for OTA. All contaminated samples remained below legislated maximum levels in the EU. In addition, near-infrared hyperspectral imaging (NIR-HSI) was applied for single-kernel classification in peanuts based on aflatoxin presence (this part was restricted to peanuts because only one positive sample of raw/roasted pistachios was available). Artificial neural network (ANN) models coupled with Principal Component Analysis (PCA) showed a stable performance, with second-derivative preprocessing yielding the best results. The most influential spectral bands were located mainly between 1100 and 1350 nm. The performance of the proposed ANN was benchmarked against three established chemometric methods: Support Vector Machines (SVM), Random Forest (RF), and Partial Least Squares Discriminant Analysis (PLS-DA). The use of naturally contaminated samples at trace levels represents a realistic advance over studies based on spiked kernels. Although the limited number of positive samples and their heterogeneous distribution present challenges for sensitivity, the findings provided evidence of the potential of NIR-HSI as a rapid, non-destructive tool to complement chromatographic methods. Future validation with larger datasets and integration into industrial sorting systems could contribute to improved routine monitoring and safer nut supply chains.