基于气相色谱法结合化学计量学识别微量 植物油的比较研究Comparative study on the identification of trace vegetable oils based on gas chromatography combined with chemometrics

为准确识别法庭科学领域中微量油脂物证，给涉及微量植物油物证鉴定的相关案件提供技术支持，以遗留在不同载体上并在4、25、38 ℃下分别放置1、3、7、14、30、45、60 d的8种微量植物油（亚麻籽油、油茶籽油、菜籽油、玉米油、花生油、芝麻油、大豆油、葵花籽油）为研究对象，利用气相色谱技术测定其脂肪酸组成，以 5种主要脂肪酸（十六烷酸、硬脂酸、油酸、亚油酸、亚麻酸）作为识别指标，结合化学计量学方法构建Fisher判别分析、卷积神经网络和随机森林3种植物油识别模型。结果表明：Fisher判别分析、卷积神经网络和随机森林3种模型均能实现对8种植物油的准确识别，其中随机森林模型能评估各脂肪酸对分类结果的重要性，且识别准确率最高，达98.2%。综上，随机森林模型参数设置简单，识别准确率高，有效解决了微量植物油种类识别困难的问题。In order to accurately identify trace oil evidence in the field of forensic science, and provide technical support for relevant cases involving the identification of trace amounts of vegetable oil evidence, eight trace vegetable oils (flaxseed oil, oil-tea camellia seed oil, rapeseed oil, corn oil, peanut oil, sesame seed oil, soybean oil, and sunflower seed oil) left on different carriers and stored at 4, 25, 38 ℃ for 1, 3, 7, 14, 30, 45 d, and 60 d respectively were used as research object. The fatty acid composition was determined by gas chromatography, and five main fatty acids (hexadecanolic acid, stearic acid, oleic acid, linoleic acid, and linolenic acid) from eight vegetable oils were selected as identification indicators to construct three vegetable oil recognition models (Fisher discriminant analysis, convolutional neural network, and random forest) using chemometrics methods. The results showed that Fisher discriminant analysis, convolutional neural network, and random forest models could all achieve accurate recognition of eight vegetable oils, among which the random forest model could evaluate the importance of each fatty acid to the classification results, and the recognition accuracy was the highest, reaching 98.2%. In conclusion, the random forest model has simple parameter settings and high recognition accuracy, and can effectively solve the problem of difficult identification types of trace vegetable oils.