Wang Yingge, Li Mengke, Xu Li, Gao Chun, Wang Cheng, Xu Lu, Jiang Shaotong, Cao Lili, Pang Min
School of Food and Bioengineering, Hefei University of Technology, Hefei 230009, China.
Anhui Jiexun Optoelectronic Technology Co., Ltd., Hefei 230000, China.
Foods. 2025 Jun 22;14(13):2186. doi: 10.3390/foods14132186.
This study explores the application of near-infrared (NIR) spectroscopy combined with machine learning for the non-destructive detection of aflatoxin in peanuts contaminated by (). The key innovation lies in the development of an optimized spectral processing pipeline that effectively overcomes moisture interference while maintaining high sensitivity to low aflatoxin concentrations. NIR spectra were collected from peanut samples at different incubation times within the spectral range of 950 to 1650 nm. Spectral data were preprocessed, and Competitive Adaptive Reweighted Sampling (CARS) selected ten characteristic bands. Correlation analysis was performed to examine the relationships between physicochemical properties, characteristic bands, and aflatoxin content. Three machine learning models-Backpropagation Neural Network (BPNN), Support Vector Machine (SVM), and Random Forest (RF)-were used to predict aflatoxin levels. The SNV-SVM model demonstrated superior performance, achieving calibration metrics (R = 0.9945, RMSE = 9.92, RPD = 14.59) and prediction metrics (R = 0.9528, RMSE = 19.58, RPD = 7.01), along with leave-one-out cross-validation (LOOCV) results (R = 0.9834, RMSE = 11.20). The results demonstrate that NIR spectroscopy combined with machine learning offers a rapid, non-destructive approach for aflatoxin detection in peanuts, with significant implications for food safety and agricultural quality control.
本研究探索了近红外（NIR）光谱结合机器学习在无损检测受（）污染花生中黄曲霉毒素的应用。关键创新在于开发了一种优化的光谱处理流程，该流程能有效克服水分干扰，同时对低浓度黄曲霉毒素保持高灵敏度。在950至1650纳米光谱范围内，于不同孵育时间从花生样品中采集近红外光谱。对光谱数据进行预处理，竞争性自适应重加权采样（CARS）选择了十个特征波段。进行相关性分析以检验物理化学性质、特征波段和黄曲霉毒素含量之间的关系。使用三种机器学习模型——反向传播神经网络（BPNN）、支持向量机（SVM）和随机森林（RF）——来预测黄曲霉毒素水平。SNV - SVM模型表现出卓越性能，实现了校准指标（R = 0.9945，RMSE = 9.92，RPD = 14.59）和预测指标（R = 0.9528，RMSE = 19.58，RPD = 7.01），以及留一法交叉验证（LOOCV）结果（R = 0.9834，RMSE = 11.20）。结果表明，近红外光谱结合机器学习为花生中黄曲霉毒素检测提供了一种快速、无损的方法，对食品安全和农业质量控制具有重要意义。
