| 史庆瑞,马泽亮,周智,贺宪权,王志强,马云霞.基于电子舌和模式识别的中成药品辨识方法研究[J].电子测量与仪器学报,2017,31(7):1081-1090 |
| 基于电子舌和模式识别的中成药品辨识方法研究 |
| Research on Chinese patent medicine identification method based on electronic tongue technology and pattern recognition |
| |
| DOI：10.13382/j.jemi.2017.07.014 |
| 中文关键词: 虚拟仪器 电子舌 离散小波变换 主成分分析 聚类分析 BP神经网络 |
| 英文关键词:virtual instrument electronic tongue discrete wavelet transform principal component analysis cluster analysis back propagation neural network |
| 基金项目:山东省自然科学基金 ( 2015CM016)资助项目 |
|
|
| Author | Institution |
| Shi Qingrui | College of Computer Science and Technology, Shandong University of Technology, Zibo 255000, China |
| Ma Zeliang | College of Computer Science and Technology, Shandong University of Technology, Zibo 255000, China |
| Zhou Zhi | Shandong Hospital of Chinese Traditional and Western Medicine, Zibo 255026, China |
| He Xianquan | College of Computer Science and Technology, Shandong University of Technology, Zibo 255000, China |
| Wang Zhiqiang | College of Computer Science and Technology, Shandong University of Technology, Zibo 255000, China |
| Ma Yunxia | Shandong Changguo Hospital, Zibo 255000, China |
|
| 摘要点击次数: 2853 |
| 全文下载次数: 11577 |
| 中文摘要: |
| 为了实现不同中成药品的快速区分辨识，采用基于虚拟仪器技术的伏安电子舌系统对治疗感冒病症的4种不同中成药品进行了检测分析。分别采用特征点提取（FPE）法和离散小波变换（DWT）法对电子舌输出信号进行预处理，以样本点的聚集性和分类效果为依据，确定较佳的特征提取方法为以db4为母小波进行的8层离散小波变换。在此基础上，分别采用主成分分析法（PCA）、聚类分析法（CA）和BP神经网络（BPNN）对不同中成药品进行区分辨识。结果表明，PCA结果中PC1和PC2累计贡献率为95.6%，除羚羊感冒片和银翘解毒片有重叠趋势外，其余各类得到有效区分；CA能够有效地观察出4种中成药品之间的差异程度，但4种药品最终被分成两类，区分效果较差；非线性分类模型BPNN对不同中成药品区分效果较好。通过优化实验，分别确定了模型的训练算法、激活函数和隐含层节点数目等参数，测试集验证表明，BPNN模型对4种中成药品的分类正确率达到100%。本研究结果可为中成药品的非感官质量评价和快速辨识研究提供技术参考。 |
| 英文摘要: |
| An electronic tongue system based on virtual instrument technology was developed and used todistinguish andanalyze 4 kinds of Chinese patent medicines in the treatment of cold symptom. The respond signal of electronic tongue was first preprocessed by the feature point extraction (FPE)method and discrete wavelet transform (DWT)method, respectively. According to clustered property and classification results of sample points, the DWT applied “db4”as mother wavelet and decomposed 8 levels was selected as a recommended feature extraction method.The principal component analysis(PCA),cluster analysis (CA)and back propagation neural network (BPNN) were then used to distinguish and identify the 4 kinds of Chinese patent medicines.The results showed that the cumulative contribution rate of PC1 and PC2 was reached 95.6% when PCA was employed.All medicines were effectively distinguished except that Lingyang cold tablet and Yinqiao antidotal tablet had an overlapping trend. The CA could obviously observe the dissimilarity of the 4 kinds of Chinese patent medicines, but the classification result of CA was so poor that 4 kinds of medicines were classified into 2 groups, while nonlinear model BPNNexhibited a better result than other classification model. The parameters of BPNN such as the training algorithm, the activation function and the number of hidden layer nodes were optimized and determined for improving the model performance. The validation set results indicate that all samples are perfectly discriminated by BPNN with the correct classification rate reaching 100%. This research can provide a technical reference for the research on non sensory quality evaluation and identification of Chinese patent medicines. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
