Multi-class Support Vector Machine for On-line Spectral Quality Monitoring of Tobacco Products

作者：Tan C^*; Wu T; Qin X

来源：Asian Journal of Chemistry, 2013, 25(7): 3668-3672.

DOI：10.14233/ajchem.2013.13712

• Quality control
• Tobacco
• Near-infrared spectroscopy
• Chemometrics

The combination of on-line near-infrared spectroscopy and chemometrics methods i.e., three kinds of multi-class support vector machine (SVM), namely, BSVM, one-against-one support vector machine (OAOSVM) and one-against-all support vector machine (OAASVM), was explored for monitoring the quality of tobacco products. The influence of the training set size on the performance was also investigated. A total of 165 samples from a cigarette factory were used for simulation. To compare different algorithms, three performance criteria were defined. The results revealed that as a whole, BSVM shows the best performance, especially in situations where the training set is comparatively small, while OAOSVM and OAASVM make no difference. Also, BSVM bears least computational cost since it can often build a classifier With less support vectors by only solving a single optimization. It seems that that BSVM could be a powerful tool for quality control based on high-dimensional spectral information.

• 出版日期2013-7
• 单位宜宾学院