融合BPSO最优粒子和被选概率特征有用光谱的选择  

Selection of NIR Spectra Combining the Optimization Particle with Probability Features of BPSO

作　　者：马孝腊 申金媛[1] 刘润杰[1] 穆晓敏[1] 

机构地区：[1]郑州大学信息工程学院,郑州450001

出　　处：《湖北农业科学》2016年第1期205-207,247,共4页Hubei Agricultural Sciences

基　　金：河南省烟草公司科技计划项目(M201335)

摘　　要：为提高基于红外光谱(NIR)对烟叶分级的效率,提出融合BPSO最优粒子和被选概率特征对烟叶的NIR进行有用特征光谱的选择。基于该方法选择的有用NIR特征光谱,对2012年642片(13个等级)烟叶进行分级,试验结果表明,通过合适的被选概率值可以得到数目相对少的用于烟叶自动分级的特征光谱组合,若与最佳粒子融合可以得到更好的分级吻合率。利用选择后的特征光谱不仅可以提高分级速度,还可以适当提高分级正确率。Aiming to improve the grading efficiency of tobacco by near infrared spectroscopy （NIR）, the selecting method combining the optimization particle with the probability features of BPSO was proposed to choose the useful characteristic spectrum of tobacco NIR. 642 pieces of tobacco leaves（including 13 grades） collected in 2012 were graded based on the useful features chosen by the presented selection method, the results indicated that, the numbers of feature spectrum used to grade the tobacco leaves automatically was decreased highly by determining the suitable values of selected probability. The better correct grading rate could be obtained when combining the features selected by probability with the optimum particle features.Experiment results showed that features chosen by BPSO could not only improve the classification speed, but also raise the correct grading rate.

关 键 词：近红外光谱 BPSO 被选概率 支持向量机 

分 类 号：TN219[电子电信—物理电子学]