红外光声光谱技术结合支持向量机鉴别油菜籽品种

Classification of rapeseed varieties using fourier transform mid-infrared photoacoustic spectroscopy(FTIR-PAS) combined with support vector machines(SVM)

采用傅里叶变换红外光声光谱技术对10个品种的油菜籽样本进行品种鉴别。原始光声光谱卷积平滑后,首先采用全谱数据建立支持向量机鉴别模型,当RBF核函数的核参数γ值为0.01时,模型最大预测率为70%。利用方差分析的方法对全谱进行有效波长筛选,筛选后的波长用于建立支持向量机鉴别模型,当γ值取0.1时,模型的识别率和预测率均可达到100%。同时,采用偏最小二乘判别分析建立鉴别模型,作为支持向量机模型的对照,该模型的预测率仅为60%,明显低于支持向量机模型的预测精度。研究表明,红外光声光谱技术结合支持向量机,在油菜籽品种鉴别中有良好的应用性能。

Fourier transform infrared phtoacoustic spectroscopy （FTIR-PAS） combined with support vector machines （SVM）, was employed to classify l0 varieties of rapeseeds. The obtained spectra smoothed by a Savitky-Golay algorithm were used as the input of SVM models. The maximal prediction rate was 70 % when the core parameter of RBF core function, T was set to 0.01. When characteristic wavelengths selected by variance analysis were employed to develop the SVM model and the T equaled 0.1, both the rates of recognition and prediction reached 100 %. Meanwhile, partial least squares-discriminant analysis （PLS-DA） was used to build a classification model as a comparison with SVM models. The prediction rate achieved by PLS-DA was merely 60 %, far lower compared to SVM models. This study has shown that FTIR-PAS combined with SVM hold a good promise in the classification of rapeseed varieties.