本文已被:浏览 1242次 下载 572次 |
![本文二维码信息](/download_upload_file.aspx?file_name=/uploadfile/two_dimensional_code_img/zgnydxxb/2020/1/20200108.png) 码上扫一扫! |
东北水稻叶片SPAD遥感光谱估算模型 |
张新乐1,于滋洋1,李厚萱1,刘焕军1*,张忠臣2,赵明明1,王翔1
|
|
(1.东北农业大学 资源与环境学院, 哈尔滨 150030;2.东北农业大学 农学院, 哈尔滨 150030) |
|
摘要: |
为通过构建高精度SPAD遥感估算模型,实现对水稻叶片叶绿素含量进行实时无损的监测,以东北地区多时期不同施氮水平下水稻叶片光谱反射率为研究对象,采用回归模型与BP神经网络算法构建不同输入量的SPAD高光谱估算模型,通过模型精度评价指标决定系数R2、均方根误差RMSE,确定最优输入量和最优模型。结果表明:1)不同品种水稻成熟时期不同导致在孕穗期和抽穗期之间光谱反射率出现差异;2)回归模型中以DVI(D755,D930)为变量建立多项式模型估算精度最高;3)与回归模型相比,不同波长处单波段反射率作为输入量的BP神经网络模型估算精度显著提高,R2为0.98。BP神经网络模型在隐藏节点数为7时估算精度达到稳定,在可见光和近红外处经过不同波段反射率作为输入量的尝试说明神经网络模型较为稳定,可以用来反演叶绿素相对含量。 |
关键词: 水稻 高光谱 神经网络 叶绿素相对含量 植被指数 一阶微分光谱 |
DOI:10.11841/j.issn.1007-4333.2020.01.08 |
投稿时间:2018-10-17 |
基金项目:国家重点研发计划(2016YFD0300604-4);国家自然科学基金(41671438) |
|
Remote sensing estimation model of SPAD for rice leaves in Northeast China |
ZHANG Xinle1,YU Ziyang1,LI Houxuan1,LIU Huanjun1*,ZHANG Zhongchen2,ZHAO Mingming1,WANG Xiang1
|
(1.College of Resources and Environment, Northeast Agricultural University, Harbin 150030, China;2.Agricultural College, Northeast Agricultural University, Harbin 150030, China) |
Abstract: |
To realize the real-time and lossless monitoring of chlorophyll content in rice leaves by using high precision SPAD remote sensing estimation model, the spectral reflectance of rice leaves under different nitrogen application levels in Northeast China was taken as study object. A SPAD hyperspectral estimation model with different input amounts was constructed by using regression model and BP neural network algorithm. The optimal input quantity and the optimal model were determined by the model precision evaluation index determination coefficient(R2)and the root mean square error(RMSE). The results showed that: 1)The spectral reflectance of different rice varieties was different at booting stage and heading stage; 2)The polynomial model based on DVI(D755, D930)was the most accurate among the regression models. 3)Compared with the regression model, the estimation accuracy of the BP neural network model with single band reflectivity as input at different wavelengths was significantly improved and its R2 was up to 0. 98. The estimation accuracy of BP neural network model was stable when the number of hidden nodes was 7. It was proved that the neural network model was more stable and could be used to invert the relative chlorophyll content by using different band reflectivity as input at visible and near infrared. The results illustrated the spectral response mechanism of different rice varieties in Northeast China, and provided a technical method for the high-precision inversion of SPAD in rice leaves and the regulation of normal growth process of rice in Northeast China. |
Key words: rice hyperspectral neural network relative chlorophyll content vegetation index first order differential spectroscopy |