激光诱导荧光结合AdaBoost算法的食用油分类

Classification of edible oil by laser⁃induced fluorescence combined with AdaBoost algorithm

为了快速、准确地识别食用油类型,采用激光诱导荧光技术,利用405 nm激光在食用油表面诱导离子体激发并发出荧光,由SpectraSuite软件采集和记录荧光光谱数据。以5种食用油为实验对象,实验采集的光谱数据共750组,分别采用4种不同的算法(BP算法、AdaBoost⁃BP算法、AdaBoost⁃DT算法、AdaBoost⁃KNN算法)对光谱数据建立训练模型,通过4种模型的比较得出,以KNN算法作为弱分类器的自适应提升(AdaBoost)模型对光谱数据的分类效果最好,迭代100次后测试准确率为100%,且泛化误差为0,表明此算法具有很好的泛化性能和稳定性。

In order to identify the type of edible oil quickly and accurately,the laser⁃induced fluorescence technology is adopted,that is,the 405 nm laser is used to induce and excite fluorescence on the surface of edible oil.The fluorescence spectrum data is collected and recorded by SpectraSuite software.Five kinds of edible oil were taken as the experimental objects,and 750 groups of spectral data were collected in the experiment.Four different algorithms(BP algorithm,AdaBoost⁃BP algorithm,AdaBoost⁃DT algorithm,AdaBoost⁃KNN algorithm)are adopted to establish the training model of spectral data.By comparison of the four models,it is concluded that the KNN algorithm used as the AdaBoost model of weak classifier is the best for classification of the spectral data.The testing accuracy is 100%after 100 iterations,and the generalization error is 0,which indicate that it has good generalization performance and stability.