返回列表
基于彩色图像的可见光-近红外高光谱识别牛肉产地
食品检测
2026/09/21
摘要：高光谱成像技术在肉类产地鉴别中具有潜在应用价值。然而，设备成本高、成像时间长以及高光谱图像分辨率低 等问题，限制了其在肉类在线检测中的广泛应用。本研究通过比较多种光谱预处理、特征波长提取和建模方法，开发了 一种基于可见-近红外（（ Vis-NIR）高光谱的牛肉产地鉴别方法。此外，本研究首次提出了一种结合迁移学习的多阶段渐进 式图像恢复网络（Tran-MPRNet），用于从彩色图像（RGB）中重建 Vis-NIR 高光谱数据，以实现牛肉产地的快速鉴别。 实验结果表明，使用MPRNet进行光谱重建的平均绝对误差（（ MAE）为0.0236，而通过Tran-MPRNet进一步将MAE降 低至0.00986。该模型无需依赖光谱响应函数即可构建光谱重建模型，克服了当前光谱重建模型的局限性。利用重建的光 谱数据，通过卷积神经网络进行牛肉产地分类，准确率达到91.01%。本研究不仅展示了Tran-MPRNet的潜在应用价值， 还为光谱重建技术在快速、低成本肉类检测中的应用提供了理论框架。
关键词：光谱重建；高光谱成像；彩色图像；肉类检测；地理溯源
期刊：Journal of Food Composition and Analysis
DOI：https://doi.org/10.1016/j.jfca.2024.106622
设备：GaiaField
Identiffcation of geographical origin of beef by visible-near infrared hyperspectral recovery from color images
Abstract：Hyperspectral imaging technology has potential in meat origin authentication. However, the high device costs, long imaging time and low hyperspectral image resolution are major obstacles limiting its wider applications in meat online detection. By comparing various spectra preprocessing, feature wavelength extraction and modeling methods, an origin identiffcation method for beef based on visible-near infrared (Vis-NIR) hyperspectra was developed. A multi-stage progressive image restoration (MPRNet) combined with transfer learning (TranMPRNet) for reconstructing Vis-NIR hyperspectra from color images (RGB) was proposed for the ffrst time in this study to identify the origin of beef. The results showed that utilizing the MPRNet achieved a mean absolute error (MAE) of 0.0236. Furthermore, the MAE was further reduced to 0.00986 through our proposed Tran-MPRNet. This model allowed the spectral reconstruction model to be built without spectral response functions, overcoming the current limitation of spectral reconstruction models. Spectra reconstructed from RGB were employed for beef origin classiffcation by a convolutional neural network, achieving an accuracy of 91.01 %. This research not only highlights the potential applications of Tran-MPRNet, but also provides a theoretical framework for the use of spectral reconstruction in rapid and cost-effective detection of meat.
Keywords：Spectral reconstruction; Hyperspectral imaging; Color images; Meat detection; Geographical origin
