论文部分内容阅读
以不同产地和年份的农华101(NH101)玉米杂交种和母本种子为对象,研究了鉴别玉米杂交种子纯度的近红外光谱分析方法。光谱采集时间跨度达10个月,运用傅里叶变换(FT)近红外光谱仪器,在不同季节用23天(分五个时间段)采集了这些样品共920条玉米单子粒近红外漫反射光谱。全部原始光谱用移动窗口平均、一阶差分导数和矢量归一化进行预处理,使用主成分分析(PCA)方法和线性判别分析(LDA)方法降维,采用仿生模式识别(BPR)方法建立模型。通过对光谱预处理校正光谱失真,使样品光谱集在特征空间分布的范围收缩,相对距离增大了近70倍,实现了母本和杂交种子的鉴别。通过代表性样品的选择,提高了模型对光谱采集时间、地点、环境等条件变动的应变能力,也提高了模型对样品种子制种时间与地点变动的应变能力,增强了模型的稳健性,使测试集玉米单子粒杂交种和母本种子的平均正确识别率达到95%以上,而平均正确拒识率也达到85%以上。
In this paper, Nanyuan 101 (NH101) maize hybrids and maternal seeds in different producing areas and years were used to study the near-infrared spectral analysis method for identifying the purity of maize hybrid seeds. The spectral acquisition time span of 10 months, the use of Fourier transform (FT) near infrared spectroscopy, in different seasons with 23 days (five time periods) were collected a total of 920 samples of corn single grain near-infrared diffuse reflectance spectroscopy . All original spectra were preprocessed by moving window averaging, first-order difference derivative and vector normalization. Principal component analysis (PCA) method and linear discriminant analysis (LDA) method were used to reduce the dimension of the original spectrum. BPR was used to establish the model . By correcting the spectral distortion by spectral preprocessing, the spectrum of the sample is shrank in the range of the characteristic spatial distribution, and the relative distance is increased by nearly 70 times. The identification of the female parent and the hybrid seed is achieved. Through the selection of representative samples, the ability of the model to respond to changes in time, location, environment and other conditions of the spectral acquisition is improved, the model’s ability of adapting to changes in time and place of seed seed production is enhanced, and the robustness of the model is improved The average correct recognition rate of maize single hybrids and maternal seeds in the test set was more than 95%, while the average correct rejection rate was over 85%.