Kuang Lingling, Tian Xuecong, Su Ying, Chen Chen, Zhao Lu, Ma Xuan, Han Lei, Chen Cheng, Zhang Jianjie
College of Computer Science and Technology, Xinjiang University, Urumqi 830046, China.
College of Software, Xinjiang University, Urumqi 830046, China.
Spectrochim Acta A Mol Biomol Spectrosc. 2025 Apr 5;330:125604. doi: 10.1016/j.saa.2024.125604. Epub 2024 Dec 30.
As a natural oil, horse oil has unique biological activity ingredients and therapeutic characteristics, which has important application value and market potential in healthcare, food, skin care and other fields. However, fraud is rampant in the horse oil market, and traditional methods such as chemical analysis and physical property detection are time-consuming, costly, and have low accuracy in detecting adulteration. Excessive adulteration may cause health risks, skin problems, and economic losses. Therefore, it is urgent to establish a rapid method for identifying adulteration in horse oil. Infrared spectroscopy exhibits substantial potential within detection applications, attributable to its fast analysis speed, non-destructive, and easy operation. This study collected four types of samples: horse oil, butter, sheep oil, and lard, and mixed them in different proportions (5%, 10%, 20%, 30%, 40%, 50%). The infrared spectral data were enhanced by Gaussian white noise and preprocessed by Standard normal variable transformation and detrending (SNV-DT), and 591 × 3601 infrared spectral data were obtained for each adulteration ratio. In terms of model selection, by comparing CNN, RNN, Transformer, and ResNet, which are commonly used in foods, cosmetics and other fields, it is found that the fine-tuning ResNet can achieve the best results in identifying adulterated horse oil applications. For the first time, this study proposed a method for rapid detection of horse oil adulteration by combining infrared spectroscopy and deep learning, which reflected the significance of combining deep learning and infrared spectroscopy in the field of adulteration, and laid a foundation for qualitative detection in this field.
马油作为一种天然油脂，具有独特的生物活性成分和治疗特性，在医疗保健、食品、护肤品等领域具有重要的应用价值和市场潜力。然而，马油市场上掺假现象猖獗，化学分析和物理性质检测等传统方法耗时、成本高，且在检测掺假方面准确性较低。过度掺假可能导致健康风险、皮肤问题和经济损失。因此，迫切需要建立一种快速鉴定马油掺假的方法。红外光谱在检测应用中具有巨大潜力，因其分析速度快、无损且操作简便。本研究收集了马油、黄油、羊油和猪油四种类型的样品，并将它们按不同比例（5%、10%、20%、30%、40%、50%）混合。红外光谱数据通过高斯白噪声增强，并采用标准正态变量变换和去趋势化（SNV-DT）进行预处理，得到了每个掺假比例下591×3601的红外光谱数据。在模型选择方面，通过比较食品、化妆品等领域常用的卷积神经网络（CNN）、循环神经网络（RNN）、Transformer和残差网络（ResNet），发现微调后的ResNet在识别掺假马油应用中能取得最佳效果。本研究首次提出了一种将红外光谱与深度学习相结合的马油掺假快速检测方法，体现了深度学习与红外光谱在掺假领域结合的重要性，为该领域的定性检测奠定了基础。
