基于高光谱成像技术的小麦籽粒品种鉴别方法研究

为了更精确地鉴别小麦品种,实现小麦品种快速、无损、有效、稳定的鉴别。利用高光谱成像系统采集6个小麦品种籽粒光谱和图像信息,提取小麦籽粒胚、胚乳、胚和胚乳混合部位的光谱,采用不同的预处理方法对原始光谱进行处理,利用竞争性自适应重加权算法(CARS)和连续投影算法(SPA)提取特征波长,基于全波长和特征波长建立线性判别分析(LDA)、支持向量机(SVM)和K最邻近(KNN)模型,筛选出最佳的籽粒部位光谱、预处理方法和特征波长提取方法；在此基础上,分析光谱信息、形态特征及二者结合信息对小麦品种的鉴别效果。结果表明,基于34个特征波长光谱信息结合形态特征建立的LDA模型效果最佳,其训练集和预测集的正确判别率分别为91.3%和86.0%。基于高光谱成像技术进行小麦品种鉴别是可行和有效的。

Research on Identification Method of Wheat Grain Varieties Based on Hyperspectral Imaging Technology

In order to identify wheat varieties more accurately, to identify the wheat varieties in a rapid, non-destructive, effective and stable way, hyperspectrum and image information were combined to identify the wheat varieties. 6 varieties wheat kernels were collected, and the spectrum of wheat kernels embryo, endosperm, embryo and endosperm mixed parts were collected from a hyperspectral imaging system. The original spectrums were processed using different pretreatment methods. Competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) were applied to extract feature wavelengths. Classification models were developed based on full wavelengths and feature wavelengths using linear discriminant analysis (LDA), support vector machine (SVM) and K nearest neighbor (KNN) models. The best spectrum of the kernel part, pretreatment method and feature wavelength extraction method were screened out. The influences of the spectral information, morphological characteristics information and the combination of the two information on the identification of wheat varieties were also analyzed. The results showed that the LDA model based on spectrums of 34 feature wavelengths combined with morphological characteristic had the best effect, and the overall correct discrimination rates of its training set and prediction set were 91.3% and 86%, respectively. Identification of wheat varieties based on hyperspectral imaging technology is feasible and effective.