基于无信息变量消除法的水稻种子发芽率测定

Study on germination rate of rice seed based on uninformation variable elimination method

为了解决常规的水稻种子发芽率测定方法存在的试验周期长且操作繁琐等问题,实现水稻种子发芽率的快速检测。本研究以黑龙江省五常市五优稻四号粳稻种子为研究对象,首先将7组种子样本(每组60个样本,共计420个样本)置于温度为45℃,湿度为90%的环境中分别进行为期0d、1d、2d、3d、4d、5d、6d的不同时段的人工老化,然后采集每个水稻种子样本的光谱数据后进行发芽试验。对光谱数据使用蒙特卡洛交叉验证法进行异常样本剔除,并应用UVE法对全光谱数据进行特征波长选择,使光谱数据由全光谱的1845个数据点缩减为524个数据点,最后建立PLSR预测模型。所建模型的预测集决定系数R 2为0.8170、RMSEP为2.1830。试验结果表明,经UVE法降维后建立PLSR模型的各项参数均优于全光谱模型,因此,UVE特征波长选择算法为提高水稻种子发芽率测定模型的预测能力提供了一种新的途径。

In order to solve the problems of long test period and complicated operation of conventional methods for determining germination rate of rice seeds,a method of rapid detection of germination rate of rice seeds was put forward.In this study,the seeds of japonica rice Wuyou No.4 in Wuchang City of Heilongjiang province were taken as the research object.Firstly,seven groups of seed samples(60 samples in each group and 420 samples in total)were placed at a temperature of 45℃and humidity of 90%for artificial aging at different time periods of 0 d,1 d,2 d,3 d,4 d,5 d and 6 d,respectively.Secondly,the germination experiment was carried out after collecting the spectral data of each sample of rice seeds.Monte Carlo cross validation method was used to remove abnormal samples of spectral data,and UVE method was used to select the characteristic wavelength of the whole spectrum data.The spectral data were reduced from 1845 data points to 524 data points.Finally,PLSR prediction model was established based on the 524 data points.The predictive set determination coefficient R 2 of the model was 0.8170 and RMSEP was 2.1830.The results showed that the parameters of PLSR model after dimension reduction by UVE method were better than those of full spectrum model.Therefore,the UVE characteristic wavelength selection algorithm provides a new way to improve the prediction ability of rice seed germination rate measurement model.