从识品种到辨霉变，高光谱成像赋能粮食质量全链条监控

Yang et al. (2020) 基于高光谱成像（HSI）与堆叠稀疏自动编码器（SSAE）算法，提出了一种贮藏玉米粒霉变状态的早期无损检测方法。通过对285个不同贮藏时间（0-40天）的玉米样本（依据真菌孢子数划分为健康、轻度、中度、重度霉变四类）进行高光谱数据（400-1000 nm）采集，结合SSAE网络提取光谱非线性特征，并对比传统特征选择算法（VCPA、RF）及多种分类器（KELM、ELM、SVM）构建识别模型。结果表明，SSAE-KELM模型表现最优，训练集与测试集平均分类准确率分别达97.36%和96.84%，敏感性与特异性均高于0.92。此外，研究通过像素级与物体级可视化技术直观展示了不同霉变等级在玉米粒表面的分布特征（图17）。该成果验证了深度学习方法在高光谱数据特征挖掘中的优势，为粮食仓储中早期霉变的快速在线检测提供了技术参考，并拓展了多模态特征融合在农业无损检测中的应用潜力。

图17基于像素级和对象级的玉米籽粒霉变等级可视化

Kang et al. (2022) 提出了一种基于高光谱成像技术（400–1000 nm）与协同聚类算法的无监督玉米粒霉变检测方法（图18）。通过融合模糊C均值聚类（FCM）和谱聚类（SC）开发了FCM-SC算法，有效解决了传统方法对标记数据的依赖及复杂数据分布的分类难题。研究采用标准正态变量变换（SNV）和乘性散射校正（MSC）预处理光谱数据，结合主成分分析（PCA）降维及二维Gabor纹理特征提取，实现了光谱与图像信息的协同利用。实验表明，FCM-SC算法在准确率（93.47%）、归一化互信息（0.5885）和兰德指数（0.8943）上显著优于传统聚类算法（如K-means、GMM），且在不依赖标记样本的情况下，其性能超过监督模型（如SVM、LDA）。该方法通过多阶段聚类压缩数据规模并保持非线性结构，为粮食质量无损检测提供了高效的无监督解决方案，未来可进一步优化算法实时性与泛化能力

图18 玉米籽粒霉变高光谱图像检测算法

总结与展望

高光谱成像技术结合光谱分析和图像处理的优势，实现了粮食品质的无损、快速、精准检测。近年来，该技术在多个方面取得重要进展，包括：1）成分分析，利用高光谱数据定量检测粮食中的水分、蛋白质、脂肪、淀粉等营养成分，提高品质监测的科学性；2）品质分级，结合机器学习和深度学习方法，基于光谱特征对粮食品质进行自动分级，提高分选效率；3）霉变与污染检测，通过识别霉变区域和真菌毒素污染，实现食品安全快速筛查；4）品种鉴别，基于高光谱特征提取，不同粮食品种可精准分类；5）存储和加工监测，跟踪粮食在存储、加工过程中的品质变化，如水分损失、氧化变质等，以优化储存和加工条件。

未来，高光谱成像技术在粮食品质检测领域将向智能化、便携化、多功能化方向发展。1）设备小型化与实时检测，开发便携式或在线检测系统，适用于生产线和现场检测；2）多源信息融合，结合X射线、热成像等技术，提高检测全面性和精度；3）深度学习与大数据分析，利用Transformer、CNN等优化光谱数据处理，提高计算效率和模型泛化能力；4）标准化与产业化，建立统一的光谱数据处理方法和检测标准，推动技术在粮食检测行业的广泛应用；5）光谱特征优化，研究高效波段选择和特征提取方法，降低数据冗余，提高计算速度。未来，该技术将与人工智能、物联网和自动化结合，推动粮食品质检测向高效、精准、智能方向发展，为粮食安全提供更有力的技术支持。

参考文献

An, T., Fan, Y., Tian, X., Wang, Q., Wang, Z., Fan, S., Huang, W., 2024. Green analytical assay for the viability assessment of single maize seeds using double-threshold strategy for catalase activity and malondialdehyde content. Food Chemistry. 455, 139889. https://doi.org/10.1016/j.foodchem.2024.139889.

Dhakal, K., Sivaramakrishnan, U., Zhang, X., Belay, K., Oakes, J., Wei, X., Li, S., 2023. Machine Learning Analysis of Hyperspectral Images of Damaged Wheat Kernels, Sensors.

Han, L., Tian, J., Huang, Y., He, K., Liang, Y., Hu, X., Xie, L., Yang, H., Huang, D., 2024. Hyperspectral imaging combined with dual-channel deep learning feature fusion model for fast and non-destructive recognition of brew wheat varieties. Journal of Food Composition and Analysis. 125, 105785. https://doi.org/10.1016/j.jfca.2023.105785.

Huang, H., Hu, X., Tian, J., Peng, X., Luo, H., Huang, D., Zheng, J., Wang, H., 2022. Rapid and nondestructive determination of sorghum purity combined with deep forest and near-infrared hyperspectral imaging. Food Chemistry. 377, 131981. https://doi.org/10.1016/j.foodchem.2021.131981.

Kang, Z., Huang, T., Zeng, S., Li, H., Dong, L., Zhang, C., 2022. A Method for Detection of Corn Kernel Mildew Based on Co-Clustering Algorithm with Hyperspectral Image Technology, Sensors.

Long, Y., Wang, Q., Tang, X., Huang, W., Zhang, B., 2025. Detection of starch content in maize kernel based on Raman hyperspectral imaging technique. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy. 336, 126025. https://doi.org/10.1016/j.saa.2025.126025.

Makmuang, S., Terdwongworakul, A., Vilaivan, T., Maher, S., Ekgasit, S., Wongravee, K., 2023. Mapping hyperspectral NIR images using supervised self-organizing maps: Discrimination of weedy rice seeds. Microchemical Journal. 190, 108599. https://doi.org/10.1016/j.microc.2023.108599.

Qi, H., Huang, Z., Jin, B., Tang, Q., Jia, L., Zhao, G., Cao, D., Sun, Z., Zhang, C., 2024. SAM-GAN: An improved DCGAN for rice seed viability determination using near-infrared hyperspectral imaging. Computers and Electronics in Agriculture. 216, 108473. https://doi.org/10.1016/j.compag.2023.108473.

Qiao, M., Cui, T., Xia, G., Xu, Y., Li, Y., Fan, C., Han, S., Dong, J., 2024. Integration of spectral and image features of hyperspectral imaging for quantitative determination of protein and starch contents in maize kernels. Computers and Electronics in Agriculture. 218, 108718. https://doi.org/10.1016/j.compag.2024.108718.

Song, Y., Cao, S., Chu, X., Zhou, Y., Xu, Y., Sun, T., Zhou, G., Liu, X., 2023. Non-destructive detection of moisture and fatty acid content in rice using hyperspectral imaging and chemometrics. Journal of Food Composition and Analysis. 121, 105397. https://doi.org/10.1016/j.jfca.2023.105397.

Xuan, G., Jia, H., Shao, Y., Shi, C., 2024. Protein content prediction of rice grains based on hyperspectral imaging. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy. 320, 124589. https://doi.org/10.1016/j.saa.2024.124589.

Xue, H., Xu, X., Yang, Y., Hu, D., Niu, G., 2024. Rapid and Non-Destructive Prediction of Moisture Content in Maize Seeds Using Hyperspectral Imaging, Sensors.

Yang, D., Yuan, J., Chang, Q., Zhao, H., Cao, Y., 2020. Early determination of mildew status in storage maize kernels using hyperspectral imaging combined with the stacked sparse auto-encoder algorithm. Infrared Physics & Technology. 109, 103412. https://doi.org/10.1016/j.infrared.2020.103412.

Yang, D., Zhou, Y., Jie, Y., Li, Q., Shi, T., 2024. Non-destructive detection of defective maize kernels using hyperspectral imaging and convolutional neural network with attention module. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy. 313, 124166. https://doi.org/10.1016/j.saa.2024.124166.

Zhang, L., Sun, H., Li, H., Rao, Z., Ji, H., 2021. Identification of rice-weevil (Sitophilus oryzae L.) damaged wheat kernels using multi-angle NIR hyperspectral data. Journal of Cereal Science. 101, 103313. https://doi.org/10.1016/j.jcs.2021.103313.

Zhang, L., Wei, Y., Liu, J., An, D., Wu, J., 2024. Maize seed fraud detection based on hyperspectral imaging and one-class learning. Engineering Applications of Artificial Intelligence. 133, 108130. https://doi.org/10.1016/j.engappai.2024.108130.

Zhu, H., Yang, R., Lu, M., Shi, W., Sun, W., Lv, D., Liu, H., Wu, Q., Jiang, X., Han, Z., 2025. Identification of maize seed vigor under different accelerated aging times using hyperspectral imaging and spectral deep features. Computers and Electronics in Agriculture. 231, 109980. https://doi.org/10.1016/j.compag.2025.109980.