基于无人机遥感的水稻氮素营养诊断研究

发布时间：2018-06-21 12:57

本文选题：水稻冠层 + 多光谱成像　；参考：《东北农业大学》2017年硕士论文

【摘要】：氮(N)是世界上作物生产最重要的营养之一,也是大多数农作物中含量最大的营养元素。对于水稻的生长发育来说,氮素是其不可缺少的重要营养元素之一。合理的施用氮肥不仅能提高水稻的优质与高产,还能缓解由农业生产带来的资源环境压力。由于农业生产具有很强的时效性,所以农情基本信息的获取要快速准确。本文利用无人机技术实时、快速监测水稻冠层氮素、叶绿素的营养状况,进一步指导其田间氮素的精细化管理,从而达到降低生产成本,提高氮肥利用率,减小环境污染的目的。本文以两个品种的水稻田间氮素梯度试验(N0-N5)为基础,以固定翼无人机获取水稻冠层图像为基本研究对象,从可见光遥感图像与多光谱图像中提取与地面取样点对应的图像特征值入手,分析图像特征值与地面测定值的关系,探寻水稻冠层的光谱响应特性,明确水稻冠层光谱响应的光谱参数,构建水稻冠层生理参数反演模型,由此来探究利用无人机影像对大尺度大田水稻氮素营养状况进行监测的可行性。本研究的主要结论如下:由于水稻品种对航拍水稻冠层数码影像的特征值影响较大,此次无人机拍摄的数码影像不适用于本次大田多品种水稻冠层氮素的分类与反演。相比之下,机载多光谱影像较适合大田水稻冠层氮素的分类与反演。从对不同施氮水平的单个品种水稻冠层氮素的识别结果来看,氮梯度为严重缺氮与严重施氮即N0和N5的总体分类识别结果最高,其识别率达到了95%以上,并且水稻品种与分类方法对N0与N5的总体识别精度基本没有什么较大影响。对于大田的多品种水稻氮素水平识别来说,多光谱影像经波段变换后得到的绿色归一化植被指数(GNDVI)影像识别严重缺氮与严重施氮即N0和N5的总体识别精度最高,达到93.83%,对于N0-N5所划分出的严重缺氮、微量缺氮、微量施氮、严重施氮的4个氮素等级的总体识别精度较差,最高只达到了57.47%,这也说明水稻植株氮素的微量变化在水稻冠层图像分类上表现不明显。试验结果表明,水稻冠层的光谱指数与水稻生理参数的拟合精度较高。对于两个水稻品种来说,GNDVI与水稻SPAD的相关性最大(R2分别达到0.9478与0.8587)。由此说明,用GNDVI影像进行水稻生理参数的反演监测是可行的。考虑水稻品种的影响,为了增加反演模型的普适性,将实测的两个品种水稻的数据融合后,得到整体水稻叶片SPAD与光谱指数的最优回归模型与氮素含量的最优反演模型。从而为利用水稻冠层图像特征预测水稻生理参数的实时、快速、无损监测提供参考依据。
[Abstract]:N) is one of the most important nutrients in crop production in the world, and is also the most important nutrient element in most crops. For the growth and development of rice, nitrogen is one of the important nutrient elements. The rational application of nitrogen fertilizer can not only improve the quality and high yield of rice, but also relieve the pressure of resources and environment brought by agricultural production. Agricultural production has a strong timeliness, so the basic information of agricultural conditions should be obtained quickly and accurately. In this paper, the real-time monitoring of nitrogen and chlorophyll in rice canopy was carried out by using UAV technology to further guide the fine management of nitrogen in the field, so as to reduce the production cost and improve the nitrogen utilization efficiency. The aim of reducing environmental pollution. In this paper, based on the nitrogen gradient test (N0-N5) between two varieties of paddy fields, a fixed wing UAV (Fixed-wing UAV) was used to obtain rice canopy image as the basic research object. Based on the extraction of the image eigenvalues corresponding to the ground sampling points from the visible and multispectral images, the relationship between the image eigenvalues and the ground measurements was analyzed, and the spectral response characteristics of the rice canopy were explored. The spectral parameters of the spectral response of rice canopy were determined, and the inversion model of rice canopy physiological parameters was constructed to explore the feasibility of monitoring the nitrogen nutrition status of rice in large scale field by using unmanned aerial vehicle (UAV) images. The main conclusions of this study are as follows: because rice varieties have a great influence on the characteristic values of aerial shoot rice canopy digital image, the digital image taken by UAV is not suitable for classification and inversion of canopy nitrogen in this field. In contrast, airborne multispectral images are more suitable for classification and inversion of nitrogen in rice canopy. According to the results of identification of nitrogen in the canopy of single rice varieties with different nitrogen application levels, the total classification and recognition results of N gradient was the highest for severe nitrogen deficiency and severe nitrogen application, i.e., N 0 and N 5, and the recognition rate was over 95%. And the total recognition accuracy of N0 and N5 was not greatly affected by rice varieties and classification methods. For the recognition of nitrogen level of rice varieties in the field, the green normalized vegetation index (GNDVI) image obtained from the multispectral image after band transformation was the best in the recognition of serious nitrogen deficiency and severe nitrogen application, namely, N0 and N5, the overall recognition accuracy of which was the highest, and that of N _ 0 and N _ 5 was the highest. Reached 93.83. The overall recognition accuracy of the four nitrogen grades classified by N0-N5 as severe nitrogen deficiency, trace nitrogen application and severe nitrogen application was poor. The highest value was only 57.47, which indicated that the microvariation of nitrogen content in rice canopy was not obvious in rice canopy image classification. The experimental results showed that the fitting accuracy between the spectral index of rice canopy and the physiological parameters of rice was higher. For two rice varieties, the correlation between GNDVI and Spad was 0.9478 and 0.8587 respectively. Therefore, it is feasible to use GNDVI image to invert and monitor the physiological parameters of rice. Considering the influence of rice varieties, in order to increase the universality of the inversion model, the optimal regression model of Spad and spectral index of whole rice leaves and the optimal inversion model of nitrogen content were obtained by fusing the measured data of two rice varieties. Therefore, it provides a reference for real-time, fast and nondestructive monitoring of rice physiological parameters by using the rice canopy image features.
【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S511;S127

【参考文献】