Spectral Data Modeling Based on Feature Extraction and Extreme Support Vector Regression

摘要：

Spectral data such as near-infrared spectrum and frequency spectrum can simply the modeling of the difficulty-to-measured parameters. A novel modeling approach combined the feature extraction with extreme support vector regression (ESVR) is proposed. The latent variables space based feature extraction method can successfully complete the dimension reduction and independent variable extraction. The novel proposed ESVR leaning algorithm is realized by using extreme learning machine (ELM) kernel as SVR kernel, which is used to construct final models with better generalization. The experimental results based on the orange juice nearinfrared spectra demonstrate that the proposed approach has better generalization performance and prediction accuracy.

关键词： Spectral data Feature extraction Extreme learning machine (ELM) Support vector regression (SVR) ELM kernel

作者: Liu Shaowei Yan Dong Liu Zhihua Tang Jian

作者单位: SUnit 92941, PLA, Huludao, 125001 Unit 92403, PLA, Fuzhou, 350007 Research Center of Automation, Northeastern University, Shenyang, 110004

会议类型: 国际会议

会议名称: 4th International Conference on Measuring Technology and Mechatronics Automation(第四届检测技术与机电自动化国际会议 ICMTMA 2012)

会议地点: 三亚

会议语种:英文

页码: 297-300

在线出版日期: 2012-01-06（万方平台首次上网日期，不代表论文的发表时间）