稻谷水分近红外光谱预测模型特征波长筛选  

Characteristic Wavelength Selection of Near Infrared Spectral Prediction Model of Rice Moisture

作　　者：吕都 周帅 陈中爱[1] 唐健波[1] LÜDu;ZHOU Shuai;CHEN Zhongai;TANG Jianbo(Institute of Biotechnology,Guizhou Academy of Agricultural Science,Guiyang 550006)

机构地区：[1]贵州省农业科学院生物技术研究所,贵阳550006

出　　处：《食品工业》2022年第7期320-324,共5页The Food Industry

基　　金：贵州省科技计划项目(黔科合支撑[2019]2828号);贵州省农业科学院课题(黔农科院青年基金[2019]10号)。

摘　　要：运用近红外光谱快速检测技术可以实现稻谷水分的快速检测,为减少近红外预测模型的输入变量、提升近红外模型的预测精度,采用逐步缩短波长优中选优的方法,筛选出与稻谷水分预测模型相关性高的特征波长。结果表明,采用偏最小二乘法对228个稻谷样品建立稻谷水分预测模型,近红外光谱的最佳预处理方式为消除常数偏移量。当步长为300 cm^(-1)时,筛选出的特征波长点有2007个,占全光谱的87%。将获得的特征波长进一步缩短步长至150 cm^(-1)进行划分,筛选出的特征波长点有1200个,占全光谱的52.02%,缩短步长至50 cm^(-1)进行划分,筛选出的特征波长点有550个,占全光谱的23.84%,缩短步长至10 cm^(-1)进行划分,筛选出的特征波长点有80个,占全光谱的3.47%。将最终筛选的特征波长建立稻谷水分预测模型,预测模型RCV 2为0.9781,RVAL 2为0.9700,表明仅利用全光谱3.47%的信息,就可以准确预测97.81%的样品。采用逐步缩短步长优中选优的方法,可为近红外光谱特征波长的筛选和近红外模型输入变量的减少提供技术支持。The rapid detection of rice moisture could be realized by using near infrared spectroscopy. In order to reduce the input variables and to improve the prediction accuracy of the near infrared prediction model, the characteristic wavelengths with high correlation with the prediction model of rice moisture were selected by the method of gradually shortening the step size and selecting the best. The results showed that the prediction model of rice moisture of 228 rice samples were established by partial least square method, and the best pretreatment method of near infrared spectrum was to eliminate constant offset.When the step size was 300 cm^(-1), a total of 2 007 characteristic wavelength points were selected, accounting for 87% of the total spectrum. The characteristic wavelength to be obtained were divided in steps of 150 cm, a total of 1 200 characteristic wavelength points were selected, accounting for 52.02% of the whole spectrum. Shortening the step size to 50 cm, 550characteristic wavelength points were selected, accounting for 23.84% of the whole spectrum. Shortening the step size to 10cm, a total of 80 characteristic wavelength points were selected, accounting for 3.47% of the whole spectrum. The prediction model of rice moisture was established based on the final selected characteristic wavelength, and the prediction model RCV2 was 0.978 1 and RVAL2 was 0.970 0. The method of gradually shortening the step size could provide technical support for the selection of the characteristic wavelengths of the near-infrared spectrum and the reduction of the input variables of the near-infrared model.

关 键 词：近红外光谱 特征波长 筛选 稻谷 水分 

分 类 号：O657.33[理学—分析化学] TS210.7[理学—化学]