基于自适应提升的监督局部保留投影在猪肉近红外光谱分类中的应用研究

Application of adaboost-based supervised locality preserving projection algorithm in classif ication of pork NIR spectra

借助于近红外反射光谱技术定性研究猪肉的贮藏时间,并以此进行猪肉的新鲜度评定。利用AntarisII快速傅里叶变换近红外光谱分析仪对猪肉样本进行无损检测获取光谱数据。为了克服光谱特征提取中的"小样本"问题,引入基于监督模式的局部保留投影算法(Supervised Locality Preserving Projection,SLPP),并通过结合自适应提升(AdaBoost)算法提出了基于自适应提升的监督局部保留投影算法(AdaBoost-SLPP)。实验结果表明:AdaBoost-SLPP算法通过加权联合能够显著提高单一线性特征分类算法的泛化能力,算法对于近红外光谱建立的定性预测模型能够达到100%的识别率。

With the help of near infrared reflectance spectroscopy （NIRS） technique,quantitative study of pork storage time was made to determine its freshness.Each pork sample was collected using the Antaris II FT-NIR spectrophotometer.To overcome the ＂small sample size （SSS）＂ problem in feature extraction,the supervised locality preserving projection （SLPP） algorithm was adopted in this study.Furthermore,a new feature extraction algorithm,called AdaBoost-SLPP,was proposed based on AdaBoost and SLPP algorithm.Experimental results showed that AdaBoost could improve the generalization ability of traditional liner feature extraction algorithms.For the calibration model of pork freshness,the discrimination rate has achieved 100%.