研究了不同分辨近红外光谱对汽车差速器油品牌鉴别的效果与影响.分别采集了五种不同品牌汽车差速器油在4,8,16和32 cm-1分辨率下的透射光谱.去除采集波段范围两端噪声明显部分后采用10 522.28～4 443.425 cm-1范围光谱进行分析.不同分辨率光谱的主成分分析(principal component analysis,PCA)结果表明不同品牌差速器油可以被鉴别.同时基于不同分辨率光谱全谱所建立的偏最小二乘判别分析(partial least squares-discriminant analysis,PLS-DA)以及支持向量机(support vector machine,SVM)模型所获得的判别正确率相近且均高于90%.其中以8 cm-1分辨率下PLS-DA与SVM模型的判别正确率最高.上述研究结果表明分辨率对判别分析结果的影响较小.进一步对不同分辨率下的近红外光谱采用连续投影算法(successive projections algorithm,SPA)选择特征波数,发现不同分辨率下得到的特征波数均不相同,表明分辨率影响特征波数的选择.以不同分辨率特征波数建立的PLS-DA以及SVM模型取得了较好的判别分析效果,与以不同分辨率下全谱建立的判别分析模型效果相当.研究结果表明,分辨率对差速器油品牌鉴别影响较小,而对特征波数的选择影响较大,在实际应用中应考虑到分辨率对特征波数选择的影响.总体而言,不同分辨率的近红外光谱全谱与对应的特征波数都能有效的实现对汽车差速器油品牌的鉴别。

The performance and influence of near-infrared spectroscopy with different resolutions for brand discrimination of differential oil was studied.Spectral data with resolutions of 4,8,16 and 32 cm-1 were collected,and 10 522.28～4 443.425 cm-1 spectra were analyzed after the removal of the absolute noises.The principal component analysis(PCA) of the spectral data at different resolutions indicated that the brands of differential oil could be discriminated.Partial least squares-discriminant analysis(PLS-DA) and support vector machine(SVM) models were built based on the full spectra with different resolutions.The discrimination results showed that all models obtained similar and good performances with the classification rate over 90%,and the best results were obtained when the model was built based on the full spectra with the resolution of 8 cm-1.Successive projections algorithm was applied to select effective wavenumbers,and different effective wavenumbers were selected for the spectra with different resolutions.PLS-DA and SVM models based on the selected effective wavenumbers both obtained good results.Their results were similar to the models established based on full spectra.The results indicated that spectral resolution had little effect on the discrimination results,and spectral resolution effected the selection of effective wavenumbers,and spectral resolution should be taken into consideration for the selection of effective wavenumbers in practical applications.In all,near-infrared spectroscopy with different resolutions and the corresponding selected effective wavenumbers could be used for discrimination of differential oil brands.