应用高光谱鉴别黑枸杞和唐古特白刺果  

Hyperspectra Used to Recognize Black Goji Berry and Nitraria Tanggu

作　　者：赵凡 闫昭如 宋海燕[1] ZHAO Fan;YAN Zhao-ru;SONG Hai-yan(College of Engineering,Shanxi Agricultural University,Taigu 030801,China)

机构地区：[1]山西农业大学工学院,山西太谷030801

出　　处：《光谱学与光谱分析》2021年第7期2240-2244,共5页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2018YFD0700300)资助。

摘　　要：黑枸杞含有花青素、多糖、氨基酸和微量元素等多种营养成分,具有极高的经济和医药价值,其市场价格很高。唐古特白刺果外观和黑枸杞极为相似,其价格较低,经常被用于冒充黑枸杞。高光谱图像技术结合图像和光谱于一体,常用于食品检测和识别等领域。结合高光谱图像技术,无损识别黑枸杞和唐古特白刺果。采集黑枸杞(180份)和唐古特白刺果(180份)的高光谱图像,利用掩膜提取光谱,光谱范围为900~1700 nm,共254个波段,去除前22个异常波段。采用Kennard-Stone法划分样品,校正集∶预测集=2∶1;采用连续投影算法(SPA)法对光谱进行降维,设定提取特征波长范围为0~30,最终提取特征波长为20个;分别将全光谱(FS)和SPA提取的20个特征波长作为模型输入,建立支持向量机(SVM)和极限学习机(ELM)识别模型。结果表明,基于FS和SPA建立的SVM模型识别率为100%;基于FS和SPA建立的ELM模型识别率为100%;SPA法在不降低模型识别精度的情况下,能减少模型输入,输入仅为FS的8.62%,大大降低模型运算量。此研究为识别黑枸杞和唐古特白刺果提供了参数。Black Goji berry contains various nutrients such as cyanidin,polysaccharides,trace elements and so on,and has extremely high economic and medical value,the similar Nitraria Tanggu impersonates in the market.The market price of Nitraria Tangguis low.Hyperspectral image technology combines image and spectrum in one,commonly used in food detection and recognition.This study combined with hyperspectral image technology to non-destructively identify Black Goji Berry and nitraria tanggu.Hyperspectral reflection spectra of Black Goji Berry(180)and nitraria Tanggu(180)in the range of 900~1700 nm were collected respectively,a total of 254 bands.Removing the first 22 abnormal bands and using the last 232 bands as model inputs.Kennard-Stone method is used to divide samples,correction set∶prediction set=2∶1.The successive projections algorithm(SPA)method is used for spectral dimensionality reduction,setting the characteristic wavelength range to 0~30,which extracts 20 characteristic wavelengths.The full spectrum and 20 characteristic wavelengths extracted by SPA are used as model inputs to establish support vector machine(SVM)and extreme learning machine(ELM)models to identify Black Goji Berry and nitraria Tanggu.The results show that the recognition rates of the SVM model based on FS and SPA are both 100%,the recognition rates of the ELM model based on FS and SPA are both 100%,the SPA method can reduce model input without reducing the accuracy of model recognition.The input is only 8.62%of FS,which greatly reduces the number of model calculations.This study provides a theoretical basis for identifying Black Goji Berry and nitraria Tanggu.

关 键 词：高光谱 模型 鉴别 黑枸杞 唐古特白刺果 

分 类 号：O433.4[机械工程—光学工程]