Abstract Rapid and accurate identification of edible oils is of great importance. The feasibility of the automatic classification of edible oils using convolutional neural network (CNN) and low-field nuclear magnetic resonance (LF-NMR) spectra was investigated. The classification ability of two-dimensional CNN (2D-CNN) or one-dimensional CNN (1D-CNN) using different LF-NMR spectra information were compared. The results indicated that transverse relaxation decay signals based on the 1D-CNN have the best classification ability, which can correctly classify the 11 types of edible oils within one minute. Furthermore, the reliability of this model was proved by loss rate, accuracy, sensitivity and specificity, and no over-fitting phenomenon was observed. It also was proved that the combination of CNN with LF-NMR could be an intelligent and automated approach for the classification of edible oil.

中文翻译：

---

基于卷积神经网络的低场核磁共振食用油分类方法

摘要 食用油的快速准确鉴定具有重要意义。研究了使用卷积神经网络 (CNN) 和低场核磁共振 (LF-NMR) 光谱自动分类食用油的可行性。比较了二维CNN（2D-CNN）或一维CNN（1D-CNN）使用不同LF-NMR谱信息的分类能力。结果表明，基于1D-CNN的横向弛豫衰减信号具有最好的分类能力，可以在一分钟内对11种食用油进行正确分类。此外，该模型的可靠性通过损失率、准确性、灵敏度和特异性来证明，没有观察到过拟合现象。