摘要
以吉林名贵中药材人参作为研究的主要对象,详细研究了利用小波变换技术对红外光谱变量的压缩方法和实现过程,以及如何利用人工神经网络(ANN)和支持向量机(SVM)技术建立人参的红外光谱的产地鉴别模型,并详细讨论了ANN模型中相关参数的优化方法以及SVM模型中的核函数及σ值的优化选择。仿真实验表明,建立的ANN模型对40个吉林人参样品产地识别率达到92.5%,而采用径向基核函数的SVM模型的识别率为97.5%,其分类效果明显优于ANN模型。从而表明小样本的情况下,利用SVM结合小波变换技术可以对吉林人参的红外光谱的产地特征进行正确区分,同时为中草药的红外光谱的进一步的分析和研究提供了一定理论依据和技术支持。
In the present study, 40 samples of ginsengs （20 samples from Jian and 20 samples from Fushun） were surveyed by Fourier transform infrared （IR） spectroscopy. Meanwhile, in order to eliminate the spectral differences from the baseline drifts, the original ginseng spectra were processed using first derivative method. To avoid enhancing the noise resulting from the derivative the spectra were smoothed. This smoothing was done by using the Savitzky-Golay algorithm, a moving window averaging method. Artificial neural network （ANN）, support vector machine （SVM） as the new pattern recognition technology, and wavelet transform （WT） were applied. Firstly, the spectrum variables of infrared spectroscopy were compressed through the WT technology before the models were established, in order to reduce the time in establishing models. Then, the identification models of cultivation area of ginsengs were studied comparatively by the use of ANN and SVM methods. The corresponding important parameters of models were also discussed in detail, including the parameters of wavelet compressing and training parameters of ANN and SVM models. The simulation experiment indicated that the ANN model can carry on the distinction among 40 samples of ginsengs from Jilin, and the accuracy rate of identification was 92.5%. The radial basis function （RBF） SVM classifiers and the polynomial SVM classifiers were studied comparatively in this experiment. The best experimental results were obtained using RBF SVM classifier with σ=0.6, and the accuracy rate of identification was 97.5%. Finally, compared with ANN approach, SVM algorithm showed its excellent generalization for identification results while the number of samples was smaller. The overall results show that IR speetroseopy combined with SVM and WT technology can be efficiently utilized for rapid and simple identification of the cultivation area of ginsengs, and thus provides the certain technology support and the foundation for further researching ginseng and other IR applications.
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2009年第3期656-660,共5页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金项目(50635030)
吉林省科技厅重点项目(20060902-02
200705c07)资助
关键词
红外光谱
小波变换
建模
人工神经网络
支持向量机
Infrared spectroscopy
Wavelet transform
Establishing model
Artificial neural networks
Support vector machine
