不同来路海洛因近红外光谱的支持向量机模式识别

发布时间：2018-04-15 09:27

本文选题：支持向量分类机 + 核函数　；参考：《计算机与应用化学》2009年03期

【摘要】：应用支持向量分类方法,将云南省9个地区缴获的1148个海洛因样品,用近红外漫反射光谱在4 000 cm~(-1)～10 000 cm~(-1)范围内吸收系数数据集合,构建判别毒品来路的分类器。光谱数据选取了指纹波数区段5 990 cm~(-1)～7 500 cm~(-1),以及最大和较大吸收系数的41个波数的光谱数据。针对一对一算法的五分类问题,采用两种分类法C-SVC和v-SVC, 4种核函数,分别以默认参数和优化参数,得训练集模型有效率和检验集的预报总精度。比较各种模型后,确定了152个指纹区波数,线性核函数的L-152 C-SVC作为分类器模型。该模型对已知分类的5个地区随机选取的训练集样本,在10-交叉检验下的有效率是90.74%,对不包含训练集的其余全部已知样品,其预报总精度是88.71%。5地区分类统计计算的敏感性、特异性、相关系数的评价都较好。最后,又试用该分类器于未知地毒品的来路辨认。与报道的模式识别比较,工作没有止于训练集给出模型,检验集判断预报效果的已知样品,又走出了重要一步,即识别训练集和检验集之外的未知样品。
[Abstract]:Using the support vector classification method, 1148 heroin samples from 9 regions of Yunnan Province were collected and the data of absorption coefficients were collected in the range of 4 000 cm~(-1)~10 ~ (000) cm ~ (-1) from diffuse reflectance spectra of near infrared (NIR) to construct a classifier for judging the origin of drugs.Spectral data are selected from the range of 5 990 cm~(-1)~7 / 500 cm ~ (-1) and 41 wavenumber with maximum and large absorption coefficients.In order to solve the five-classification problem of one-to-one algorithm, four kernel functions, C-SVC and v-SVC, are used to obtain the efficiency of the training set model and the total prediction accuracy of the test set by using the default parameters and the optimized parameters, respectively.After comparing various models, 152 fingerprint region wavenumber are determined, and L-152 C-SVC of linear kernel function is used as classifier model.The effective rate of this model for randomly selected training set samples from 5 regions with known classification is 90.74 under 10-cross test. For all other known samples without training set, the total prediction accuracy is 88.71.5 and the sensitivity of statistical calculation of regional classification.The specificity and correlation coefficient were all well evaluated.Finally, the classifier was used to identify the origin of the unknown drugs.Compared with the reported pattern recognition, the work does not stop at the training set to give the model, the test set to judge the prediction results of the known sample, another important step, that is, to identify the training set and test set outside the unknown samples.
【作者单位】：云南大学现代生物学研究中心;云南省公安厅物证鉴定中心;
【基金】：云南省科技攻关计划2001GG34项目资助.
【分类号】：D919

【参考文献】