Abstract:

Rape seed with high oleic acid breeding is one of the current breeding directions of rape. In order to find out an efficient and nondestructive method to analyze and determine the oleic acid content， and improve the screening efficiency of rape germplasm resources with high oleic acid， 3 rapeseed varieties were selected as materials in this study， and the spectral imaging information and oleic acid content data of seeds were collected， respectively. The spectral information was preprocessed in 11 ways， and multiplicative scatter correction CMSC was the optimal pretreatment method. 3 quantitative analysis models based on principal component analysis （PCA）， continuous projection （SPA）， competitive adaptive reweighted sampling （CARS） were established， and support vector machine （SVM）， and least square support vector machine （LSSVM） and extreme learning machine （ELM） were established to realize the nondestructive detection of rape oleic acid content. The model was tested by changing the number of training samples， and correlation coefficient （R） and root mean square error （RMSE） were used to evaluate the effect of these methods to validate the stability of the model. The results showed that MSC+CARS+ELM model had the best prediction effect among all the models. The correlation coefficient （Rc） and root mean square error RMSEc of correction set were 0.894 and 1.993 4%， respectively. The Rp and RMSEp of prediction set were 0.868 and 1.069 8%， respectively. It could more accurately predict the oleic acid content and provided theoretical basis for nondestructive testing of rapeseed nutritional quality by hyperspectral technology.

Key words: rape, hyperspectral imaging technology, oleic acid content

摘要：

高油酸油菜籽品种是当前油菜育种方向之一，为开发高效、无损测定油酸含量的方法，提高油菜高油酸种质资源筛选效率，选用3个油菜品种为材料，分别采集其种子光谱成像信息及油酸含量数据，首先对光谱信息进行11种预处理，确定多元散射校正（MSC）最佳预处理方法，然后基于主成分分析（PCA）、连续投影（SPA）、竞争性自适应重加权采样（CARS）方法对数据进行降维，最后分别建立支持向量机（SVM）、最小二乘支持向量机（LS-SVM）和极限学习机（ELM）3种定量分析模型，对油菜油酸含量进行无损检测。通过改变训练样本的数量来测试模型，为验证模型的稳定性，用相关系数（R）、均方根误差（RMSE）进行效果评价。结果表明，在所有模型中，多元散射校正+竞争性自适应重加权采样+极限学习机（MSC+CARS+ELM）模型预测效果最好，校正集相关系数（Rc）、均方根误差（RMSEc）分别为0.894、1.993 4%，预测集相关系数（Rp）为0.868，均方根误差（RMSEp）为1.069 8%，可更加准确地预测油酸含量，创建一种快速、无损检测油菜种子油酸含量的方法，为利用高光谱技术进行油菜营养品质无损检测提供理论依据。

关键词: 油菜, 高光谱成像, 油酸含量