论文部分内容阅读
将多种单分类器模型融合,并用融合后的模型对不同品种干红葡萄酒进行判别分析。用BRUKER MPA傅里叶变换型近红外光谱仪采集170个干红葡萄酒样品的近红外透射光谱,选取PLS-DA,SVM,Fisher和AdaBoost作为单分类器建模方法,分别建立葡萄酒品种判别模型,通过差异性度量值对单分类器进行筛选,得到差异性较大的四个单分类器作为基分类器,其中基分类器对测试集葡萄酒品种判别准确率最高为88.24%,最低为81.18%。然后通过加权投票机制对基分类器进行融合,融合后的模型对测试集葡萄酒品种判别准确率提高至92.94%,误判样品个数由单分类器最少的9个降为6个。实验结果表明多分类器融合所建立的模型优于传统近红外光谱定性分析一般采用单分类器模型结果,提高了葡萄酒品种判别的准确性,采用基于近红外光谱的多分类融合方法对葡萄酒种类判定具有可行性。
A variety of single-classifier models are fused and the discriminant analysis of different varieties of dry red wine is performed using the fused model. The near-infrared transmission spectra of 170 dry red wine samples were collected by BRUKER MPA Fourier transform near-infrared spectroscopy. PLS-DA, SVM, Fisher and AdaBoost were selected as single-classifier modeling methods to establish the wine variety discriminant model. According to the difference measure, the single classifier is screened to obtain four single classifiers with large difference as the base classifier. The accuracy of the base classifier for the test wine varieties is 88.24% and the lowest is 81.18%. Then, the basis classifiers are fused by the weighted voting mechanism. The fusion model improves the discrimination accuracy of test wine varieties to 92.94%, and the number of misclassified samples is reduced from 9 with a minimum number of single classifiers to 6. Experimental results show that the model established by the multi-classifier fusion is superior to the traditional single-classifier model in qualitative analysis of near-infrared spectroscopy to improve the accuracy of discriminating wine varieties. The multi-classification fusion method based on near-infrared spectroscopy Feasible.