基于消费级近红外相机的水稻叶片叶绿素（SPAD）分布问题研究

发布时间：2018-03-05 17:42

本文选题：叶绿素含量　切入点：SPAD值反演　出处：《华中农业大学》2017年硕士论文　论文类型：学位论文

【摘要】：叶绿素是作物进行光合作用有关的重要色素。实时准确的获取作物叶片的叶绿素水平,有利于掌握作物生长状况,从而确保科学的栽培和施肥,对提高作物产量,实现精准农业有着非常重要的意义。诸如SPAD仪等传统无损测量方法不能全面获取作物叶绿素含量的分布,基于光谱技术的叶绿素反演方法虽然已经日趋成熟,但叶绿素在作物体内的分布、变化却缺少可视化的表达,而及早地发现作物缺乏氮素初期引起的叶绿素性状变化,将对实际的生产提供必要的帮助。因此本研究以生长期的水稻地上部分为研究对象,提出一种基于消费级近红外相机的水稻叶片叶绿素分布获取方法,将光谱-叶绿素反演模型映射到水稻可见光和近红外图像上,得到水稻叶片叶绿素的二维分布,由此能够直观的观察水稻的生长状况,为快速、无损的低成本营养诊断提供了更多参考。主要研究内容及结果如下:本研究通过对普通单反相机加载近红外滤波片获取水稻多波段三个通道光谱信息,筛选最优近红外波段图像,建立水稻预测模型,将反演结果映射到水稻图像上,实现水稻叶绿素含量的直观描述。在筛选过程中,对比了单通道光谱信息和综合通道信息对水稻叶绿素含量预测的影响,分析了不同植被指数和拟合模型对水稻叶绿素含量预测的效果,并比较了消费级近红外单反相机成像方式和高光谱相机成像方式在叶绿素反演不同方面的优劣。结果表明,相机成像通道对于叶绿素含量的预测影响,R通道G通道B通道。以单一植被指数建立的单因子预测模型中,GVI预测精度最高,R2=0.7851,预测水稻叶绿素含量的最优波段是可见光绿光波段G和近红外760nm(NIR760R)的组合,拟合函数更接近二次曲线形式。多因子预测模型中,6个波段信息经偏最小二乘回归,R2=0.8541,4个由最优波段构造的植被指数建立的最小二乘支持向量机模型,R2=0.8314。将含有最优波段的水稻光谱信息融入叶绿素反演模型,最后得到整个水稻叶片叶面的叶绿素分布,从而实现水稻营养的准确描述、定量分析和可视化表达。总得来看,消费级近红外相机比高光谱相机更适用于在线监测。
[Abstract]:Chlorophyll is an important pigment related to photosynthesis of crops. Obtaining the chlorophyll level of crop leaves in real time and accurately is beneficial to the understanding of crop growth, so as to ensure scientific cultivation and fertilization, so as to increase crop yield. It is very important to realize precision agriculture. Traditional nondestructive measurement methods such as SPAD instrument can not obtain the distribution of chlorophyll content in crops. Although the method of chlorophyll inversion based on spectral technology is becoming more and more mature, However, the distribution of chlorophyll in crops, the changes in the lack of visual expression, and early detection of crop nitrogen deficiency caused by the initial changes in chlorophyll traits, Therefore, a method of obtaining chlorophyll distribution in rice leaves based on consumer near infrared camera is proposed. The spectral chlorophyll inversion model is mapped to the visible light and near infrared images of rice, and the two-dimensional distribution of chlorophyll in rice leaves is obtained. Thus, the growth of rice can be observed intuitively. The main research contents and results are as follows: in this study, the three channels spectral information of rice was obtained by loading NIR filter into common SLR cameras. The optimal near infrared band images were screened, the rice prediction model was established, and the inversion result was mapped to the rice image to realize the direct description of rice chlorophyll content. The effects of single channel spectral information and integrated channel information on the prediction of chlorophyll content in rice were compared, and the effects of different vegetation indices and fitting models on the prediction of chlorophyll content in rice were analyzed. The advantages and disadvantages of consumer near infrared SLR camera imaging mode and hyperspectral camera imaging mode in chlorophyll inversion are compared. The influence of camera imaging channel on the prediction of chlorophyll content in R channel G channel B channel. In the single factor prediction model established by single vegetation index, the precision of GVI prediction is the highest (R2GVI) 0.7851.The optimum band for predicting chlorophyll content in rice is that the band can be used to predict chlorophyll content in rice. See the combination of light and green band G and NIR 760nmNIR760R), The fitting function is closer to the conic form. In the multifactor prediction model, the information of 6 bands is regressed by partial least squares regression (R2N) 0.8541, and the least squares support vector machine model (LS-SVM), which is established by vegetation index constructed from the optimal band, will contain the best information. The spectral information of rice in the band is integrated into the chlorophyll inversion model. Finally, the chlorophyll distribution of the whole rice leaf was obtained, so as to realize the accurate description, quantitative analysis and visual expression of rice nutrition. In general, the consumer near infrared camera is more suitable for on-line monitoring than the hyperspectral camera.
【学位授予单位】：华中农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.33;S511

【参考文献】