分数阶微分在牛奶蛋白质光谱反演中的应用  

Application of Fractional Order Differentiation to the Spectral Inversion of Milk Proteins

作　　者：华宜赛 刘江平[1] 高凌 HUA Yisai;LIU Jiangping;GAO Ling(College of Computer and Information Engineering,Inner Mongolia Agricultural University,Hohhot 010011,China;Chifeng First Vocational Secondary Professional School,Chifeng 024000,China)

机构地区：[1]内蒙古农业大学计算机与信息工程学院,呼和浩特010011 [2]赤峰第一职业中等专业学校,赤峰024000

出　　处：《内蒙古农业大学学报（自然科学版）》2023年第3期53-59,共7页Journal of Inner Mongolia Agricultural University(Natural Science Edition)

基　　金：国家自然科学基金项目(61962048);内蒙古自治区自然科学基金项目(2022MS06026);内蒙古教育厅项目(NJZY21491);内蒙古农业大学青年教师科研能力提升专项项目(BR220116)。

摘　　要：牛奶中蛋白质含量的检测关系到牛奶品质的高低及牛奶的口感状况,试验对牛奶蛋白质含量进行研究,利用高光谱成像系统获取试验数据实现对牛奶蛋白质含量的无损检测。本文通过使用基线校正、散射校正、平滑处理和尺度缩放4个方向的8种光谱预处理算法分别对光谱进行单一方法优化,以及利用0.1阶微分为步长进行0~0.9阶分数阶微分(factional order differerntial,FOD)的数学变换,并分别使用偏最小二乘法(partial least squares,PLS)、随机森林(random forest,RF)和反向传播算法(back propagation,BP)分别对上述处理后的17种预处理结果进行建模。结果表明,采用RF的0.2阶微分建立的牛奶蛋白质含量反演模型预测精度最高,测试集的决定系数R2=0.9998,均方根误差RMSE=0.0031,相对分析误差RPD=85.1109。研究为进一步实现牛奶蛋白质含量精准检测提供了参考依据。The detection of protein content in milk is related to the quality of milk and the taste of milk,this paper aimed to study the protein content of milk by carrying experiments,and realize the nondestructive detection of milk protein content by using a hyperspectral imaging system to obtain experimental data.The spectra were optimized by a single method using eight spectral preprocessing algorithms in four directions,i.e.,baseline correction,scattering correction,smoothing and scale scaling,and mathematical transformations of fractional order differentiation(FOD)from 0 to 0.9 using the 0.1-order differentiation as the step size and modeled by partial least squares(PLS),random forest(RF)and backpropagation(BP),respectively.The partial least squares(PLS)method,random forest(RF)method,and back propagation(BP)algorithm were used to model the 17 pre-processed results.The highest prediction accuracy of the inverse model of milk protein content was established by using 0.2-order differentiation of RF,with the coefficient of determination of the test set R2=0.9998,root mean square error RMSE=0.0031,relative analysis error RPD=85.1109.It can provide a reference for the further realization of accurate milk protein content detection.

关 键 词：牛奶 分数阶微分 光谱预处理 蛋白质含量 

分 类 号：O657.3[理学—分析化学]