基于BPSO和SVM的烟叶近红外有用特征光谱选择  

Screening the effective features in the near-infrared spectroscopy of tobacco leaf based on BPSO and SVM

作　　者：李航[1] 赵海东[1] 申金媛[1] 刘润杰[1] 刘剑君 穆晓敏[1] 

机构地区：[1]郑州大学信息工程学院,河南郑州450001 [2]郑州市烟草专卖局,河南郑州450006

出　　处：《物理实验》2015年第6期8-12,共5页Physics Experimentation

基　　金：河南省烟草公司科技计划项目(No.M201335)

摘　　要：为提高基于近红外光谱识别烟叶等级的效率,利用BPSO联合SVM对原始光谱数据进行有用特征光谱选择.利用BPSO将对分级影响不好或没有影响的特征剔除,采用SVM对烟叶的等级进行识别.结果表明:BPSO选择的最佳特征光谱可减少特征光谱的数目,提高烟叶的正确分级率.对于相同的光谱范围,采样间隔越大,经过特征光谱选择后,原始光谱数据数目减少的比例越大.此外,有用特征光谱的选择可以有效地减少光谱数据的采集量,减少了分级模型的计算复杂度,提高烟叶分级的速度.To improve the classification efficiency of tobacco leaves based on near-infrared spec- troscopy, the BPSO and SVM methods were applied to screening the effective features from the origi- nal spectra. The BPSO method was used to get rid of some features that bad bad effect or no effect on the classification, and then the levels of the tobacco leaves were recognized by SVM. The experimen- tal results showed that BPSO method could greatly reduce the number of characteristic spectral data and improve the recognition efficiency. For the same spectrum range, after screening, large sampling interval could reduce the numbers of characteristic spectral data. Moreover BPSO could effectively re- duce the size of spectrum data collection and the computational complexity of the hierarchical model, thus greatly improve the classification speed.

关 键 词：近红外光谱 BPSO 支持向量机 烟叶分级 

分 类 号：TN219[电子电信—物理电子学]