太赫兹光谱在转基因菜籽油鉴别中的应用:基于改进蜉蝣算法的支持向量机模型

Application of terahertz spectroscopy in identification of transgenic rapeseed oils:A support vector machine model based on modified mayfly optimization algorithm

为实现对转基因和非转基因菜籽油的快速准确鉴别,结合太赫兹时域光谱技术,提出了一种基于改进蜉蝣优化算法的支持向量机模型.以两种转基因和两种非转基因菜籽油为研究对象,应用太赫兹时域光谱技术获取其光谱信息,发现相比于非转基因菜籽油,转基因菜籽油在太赫兹波段具有更强的吸收特性,同时它们的吸收光谱极为相似,难以通过观察法进行准确区分.为此,提出一种基于改进蜉蝣优化算法的支持向量机模型,通过采用蜉蝣优化算法对支持向量机参数进行寻优,并引入自适应惯性权重和Lévy飞行两种策略改进蜉蝣优化算法在寻优过程容易陷入局部最优解的问题,增强蜉蝣优化算法的全局搜索能力和稳健性.实验结果表明:改进后的蜉蝣优化算法能够更有效地寻找到支持向量机的最优参数组合,提升鉴别模型的整体性能,该模型对4种菜籽油的识别精度为100%.因此,本研究为转基因菜籽油的类型鉴别提供了一种快速有效的新方法,也为其他转基因物质的鉴别提供了有价值的参考.

To achieve rapid and accurate identification of genetically modified(GM)and non-GM rapeseed oils,a support vector machine(SVM)model based on an improved mayfly optimization algorithm and coupled with the terahertz time-domain spectroscopy,is proposed.Two types of GM rapeseed oils and two types of non-GM rapeseed oils are selected as research subjects.Their spectral information is acquired by using the terahertz time-domain spectroscopy.The observations show that GM rapeseed oils exhibit stronger terahertz absorption characteristics than non-GM rapeseed oils.However,their absorption spectra are highly similar,making direct differentiation difficult through visual inspection alone.Therefore,SVM is used for spectral recognition.Considering that the classification performance of SVM is significantly affected by its parameters,the mayfly optimization algorithm is combined to optimize these parameters.Furthermore,adaptive inertia weight and Lévy flight strategies are introduced to enhance the global search capability and robustness of the mayfly optimization algorithm,thus addressing the issue of easily becoming trapped in local optima in the optimization process.Moreover,principal component analysis is used to reduce the dimensionality of the absorbance data in a 0.3-1.8 THz range,aiming to extract critical features,thereby enhancing modeling efficiency and reducing redundancy in spectral data.Experimental results demonstrate that the improved mayfly optimization algorithm effectively identifies the optimal parameter combination for SVM,thereby enhancing the overall performance of the identification model.The proposed SVM model,in which the improved mayfly optimization algorithm is used,can achieve a recognition accuracy of 100%for the four types of rapeseed oils,surpassing the 98.15%accuracy achieved by the SVM model with the original mayfly optimization algorithm.Thus,this study presents a rapid and effective new approach for identifying GM rapeseed oils and offers a valuable reference for identifying other genetically modified substances.