基于高光谱成像技术小麦籽粒霉变鉴别方法研究

Identification Method of Wheat Grain Mildew Based on Hyperspectral Imaging Technology

小麦霉变籽粒是小麦不完善粒的一种,鉴别小麦霉变籽粒是粮食霉变程度的重要参考。为了更精确地鉴别小麦霉变籽粒,本研究利用高光谱成像技术采集不同品种小麦霉变籽粒及非霉变籽粒的光谱信息,建立小麦霉变籽粒的鉴别预测模型,实现小麦霉变籽粒快速、无损、有效、稳定鉴别。收集100粒霉变和100粒正常非霉变籽粒400~1000 nm范围的高光谱图谱,通过不同的光谱预处理方法进行处理,选出最优光谱信息预处理方法。采用连续投影算法(SPA)和竞争性自适应重加权采样(CARS)提取特征波长,利用偏最小二乘判别分析(PLS-DA)和支持向量机(SVM)算法分别建立全波长范围和特征波长的小麦霉变籽粒鉴别模型。结果表明,白麦和红麦的最优预测模型分别为全波长-OSC-SVM和SPA-OSC-SVM模型,对应的R^(2)_(P)分别为0.9963和0.9998,RMSEP分别为0.0309和0.0064,R^(2)_(CV)分别为0.9975和0.9995,RMSECV分别为0.0247和0.0111。针对2个品种,用SPA法挑选出的特征波长建立的SVM模型对霉变籽粒均有较好的预测效果,模型预测准确性达98%以上,满足在线快速检测准确性的要求,并且对比建模的变量数量,所利用的特征波长也相对较少。因此,最终选择SPA挑选的特征波长结合支持向量机建立霉变籽粒的鉴别模型。

Moldy wheat grains are a kind of imperfect wheat grains.The identification of moldy wheat grains is an important reference for the degree of grain mildew.In order to identify wheat moldy grains more accurately,in the present paper,hyperspectral imaging technology was used to collect the spectral information of different varieties of wheat moldy and non-mildew grains,an identification and prediction model was established for wheat moldy grains,and rapid and non-destructive wheat moldy grains were realized.First,100 grains of mildew and 100 grains of normal non-mildew grains 400~1000 nm range hyperspectral atlas were collected,processed by different spectral preprocessing methods,and the optimal spectral information preprocessing method was selected.Then,the continuous projection algorithm(SPA)and the competitive adaptive re-weighted sampling(CARS)were used to extract the characteristic wavelength,partial least squares discriminant analysis(PLS-DA)and support vector machine(SVM)algorithms were used to establish a model for identifying moldy wheat grains in the full wavelength range and characteristic wavelength.The optimal prediction models for white wheat and red wheat were the full-wavelength-OSC-SVM and SPA-OSC-SVM models,respectively.The corresponding R^(2)_(P) was 0.9963 and 0.9998,the RMSEP was 0.0309 and 0.0064,the R^(2)_(CV) was 0.9975 and 0.9995,the RMSECV was 0.0247 and 0.0111,respectively.For the two varieties,the SVM model established by the characteristic wavelength selected by the SPA method had a good predictive effect on the moldy grains,and the prediction accuracy of the model was more than 98%,meet the requirements of online rapid detection accuracy,the number of variables in the model was compared,and the characteristics used were relatively few wavelengths,so the characteristic wavelength selected by SPA was finally selected in combination with the support vector machine to establish a moldy grain identification model.