基于小波分析的水稻籽粒直链淀粉含量高光谱预测

Application of continuous wavelet analysis to laboratory reflectance spectra for the prediction of grain amylose content in rice

水稻籽粒直链淀粉含量影响稻米的蒸煮食味品质。利用遥感技术及时、准确地获取籽粒直链淀粉含量可以指导相应栽培措施的制定与实施,以提高稻米的食味品质。小波分析作为光谱敏感特征提取的有效方法,广泛应用于作物生理生化参数的估算,然而基于小波分析的作物品质参数估算,在米粉、稻穗水平上的应用还未见报道。本文以室内获取的水稻米粉与干穗反射光谱为基础数据源,通过连续小波光谱变换、敏感小波特征提取、共性特征分析和预测模型构建等4个步骤,明确不同光谱参数预测水稻籽粒直链淀粉含量的性能,最终实现在器官水平的直链淀粉含量高光谱预测。结果表明:(1)相较归一化光谱指数,敏感小波特征可有效地提高直链淀粉含量预测精度,预测模型更具普适性和鲁棒性;(2)从米粉光谱提取的敏感小波特征WF2037,6,与籽粒直链淀粉含量相关性较高(R^(2)=0.59),对独立年份的样本预测效果较好(RMSE=1.51%,Bias=0.44%,RRMSE=23.50%),并可直接应用于干穗光谱(R^(2)=0.62,RMSE=1.49%,Bias=−0.17%,RRMSE=25.76%)。本文利用连续小波光谱分析,提取了米粉和稻穗水平的直链淀粉敏感小波特征WF2037,6,建立了高精度预测模型,拓宽了连续小波光谱分析的应用范围,为冠层水平水稻籽粒直链淀粉含量的高光谱估算奠定基础。

Grain amylose content(GAC)is a critical factor affecting the cooking and eating quality of rice.Remote sensing technology can be used to obtain amylose content timely and accurately,which are useful for the establishment and implementation of corresponding cultivation to improve the quality of rice taste.As an effective method for spectral feature extraction,continuous wavelet analysis(CWA)has been widely used to estimate crop physiological and biochemical parameters.However,none of previous studies have used CWA for crop quality estimation and investigated the application of CWA to dried grain powder reflectance spectra acquired in the laboratory for absorption feature extraction.The method was conducted in four steps as below:continuous wavelet transforms,extraction of sensitive wavelet features,analysis of common features,and construction of predictive models.Finally,we estimated GAC on grain scale and compared the performance of different spectral features.The results were as follows:(1)The performance of sensitive wavelet features was better vegetation indices and the independent validation confirmed this superiority;(2)The GAC could be estimated from WF2037,6 with a high R^(2)=0.59 and the accuracy assessed with validation data from an independent year was RMSE=1.51%,Bias=0.44%,RRMSE=23.50%.The results derived from dried grain powder could be applied to dried panicles(R^(2)=0.62,RMSE=1.49%,Bias=–0.17%,RRMSE=25.76%).This study determined the optimal amylose-sensitive wavelet feature WF2037,6.It can provide new insight into GAC estimation with hyperspectral remote sensing and this method would advance the understanding of rice quality estimation from reflectance spectra at grain and canopy levels.