Abstract: Sweet corn is an important fruit and vegetable crop in the world. It is of great significance to establish a rapid and non-destructive method detecting sweet corn sugar content for the identification and screening of sweet corn materials for quality breeding. This experiment selected 104 sweet corn samples with high sugar content, and carried out near infrared spectroscopy analysis on them. Combined with the chemical value measured by enzymatic method, partial least square method and different spectral pretreatment and mathematical treatment were adopted to conduct stoichiometric analysis on the total spectral band and constructed the near infrared spectroscopy calibration model of glucose, fructose and sucrose content. External verification was carried out by sample set of external validation, and the actual prediction of the constructed model was tested. The results showed that the optimal spectral treatment of glucose was a weighted multicomponent scattering correction, fructose and sucrose were standard normalized plus descattering treatment. The optimal derivative treatment of glucose was a second derivative, fructose and sucrose were the 1st and 3rd derivatives. The correlation coefficients of glucose, fructose and sucrose were 0.646, 0.645 and 0.820, respectively. The standard deviation of interaction validation was 0.321, 0.275 and 1.508, respectively. The predicted correlation coefficients of glucose, fructose and sucrose were 0.593, 0.780 and 0.891, respectively. These results indicated that the established fructose and sucrose prediction models had better predictability and could be applied to select sweet corn germplasm resources, while the predictability of glucose model was poor and needed to be improved.

Key words: sweet corn, glucose content, fructose content, sucrose content, near infrared spectral calibration models

摘要： 甜玉米是世界上重要的果蔬两用型作物之一。建立快速、无损检测甜玉米糖分含量的方法，对于甜玉米品质育种工作中的材料鉴定、筛选具有重要意义。选取了104份糖分含量变幅较大的甜玉米材料进行了近红外光谱分析，结合酶法测量的化学值，利用偏最小二乘法以及不同光谱处理和数学处理相结合，对全光谱波段进行了化学计量学的分析统计，构建了葡萄糖、果糖和蔗糖含量的近红外光谱定标模型。用外部验证样品集进行验证，对所建模型的实际预测能力进行检验。结果表明：最优光谱处理方式葡萄糖为加权多元散射校正，果糖和蔗糖的为标准正常化加去散射处理；最优的导数处理葡萄糖为2阶导数，果糖和蔗糖的为1阶和3阶导数。葡萄糖、果糖和蔗糖定标模型的交叉验证相关系数分别为0.646、0.645、0.820，交叉验证标准偏差分别为0.321、0.275、1.508；外部验证集葡萄糖、果糖和蔗糖的预测相关系数分别为0.593、0.780、0.891，表明所建立的果糖和蔗糖预测模型具有较好的预测性，可应用于甜玉米种质资源筛选，而葡萄糖模型的预测性较差，需继续完善。

关键词: 甜玉米, 葡萄糖含量, 果糖含量, 蔗糖含量, 近红外光谱定标模型