多尺度分割的作物长势参数反演研究

发布时间：2018-09-06 09:55

【摘要】：为实现区域尺度上对作物整体长势的实时、无损监测,本文以2012年6月-7月在黑河流域开展的大型星-机-地遥感综合试验获取的CASI航空观测数据及地面实测数据为主要数据源,以eCognition影像处理软件为平台,利用多尺度分割技术对高光谱遥感影像进行信息提取,进而对长势参数LAI、Cab、N进行反演研究,旨在探讨和分析面向对象的多尺度分割对高光谱遥感作物长势参数反演的影响,以求从理论和方法上获得有关高光谱数据农业应用的一点新成果和经验。本文主要围绕以下三点进行深入探讨研究：首先,基于面向对象的多尺度分割技术,结合目视解译和逐步试错法,并在相对区域重叠度的评价指标下筛选出最优分割尺度参数,即各个波段的权重因子为1,形状因子权重为0.4,紧密度为0.5,分割尺度大小为70。其次,在综合考虑CASI影像特征和以往研究的基础上,对作物长势参数与植被指数进行统计描述及相关性分析,建立由反演能力及其稳定性均较好的植被指数组成的多元线性遥感估测模型。最后,根据面向对象多尺度分割后影像的特点,结合符合研究区实际情况的作物长势参数反演模型,对研究区以两种实验方案即影像的先分割再反演方法和先反演再分割方法进行对比分析研究,实现了区域尺度上作物整体长势的遥感监测。研究结果表明：先分割后反演的结果优于先反演后分割方法,其LAI、Cab、N的模型估测精度不仅都达到了80%以上,而且决定系数R2也均通过了0.95可靠性水平的显著性检验,充分展示了面向对象多尺度分割技术应用于作物参数反演的独特优势,使得针对整体大田作物的反演结果更加客观、准确,具有普适性,实现了高光谱遥感影像对于大面积大区域的作物长势信息的有效提取。
[Abstract]:In order to achieve real-time, non-destructive monitoring of the overall growth of crops at the regional scale, In this paper, the CASI aeronautical observation data and ground measured data obtained from the large-scale satellite-machine-ground remote sensing experiment carried out in the Heihe River Basin from June to July 2012 are taken as the main data sources, and the eCognition image processing software is used as the platform. Using multi-scale segmentation technology to extract information from hyperspectral remote sensing image, and then to study the inversion of growth parameter LAI,Cab,N. The purpose of this paper is to discuss and analyze the effect of object-oriented multi-scale segmentation on the inversion of hyperspectral remote sensing crop growth parameters. In order to obtain some new achievements and experiences about the application of hyperspectral data in agriculture in theory and method. This paper mainly focuses on the following three aspects: firstly, based on the object-oriented multi-scale segmentation technology, combined with visual interpretation and step by step trial and error method, the optimal segmentation scale parameters are selected under the evaluation index of relative regional overlap degree. That is, the weight factor of each band is 1, the weight of shape factor is 0.4, the tightness is 0.5, and the scale of segmentation is 70. Secondly, on the basis of considering the characteristics of CASI images and previous studies, the relationship between crop growth parameters and vegetation index was analyzed. A multivariate linear remote sensing estimation model composed of vegetation indices with good inversion ability and stability was established. Finally, according to the characteristics of the object oriented multi-scale segmented image and the inversion model of crop growth parameters in accordance with the actual situation in the study area, In this paper, two experimental schemes, the first segmentation and then inversion, and the first inversion and segmentation, are compared and analyzed, and the remote sensing monitoring of the whole crop growth on the regional scale is realized. The results show that the results of first segmentation and then inversion are superior to those of the first inversion and then the segmentation method. The model estimation accuracy of LAI,Cab,N not only reaches more than 80%, but also the determination coefficient R2 has passed the significance test of 0.95 reliability level. It fully demonstrates the unique advantages of object-oriented multi-scale segmentation technology in crop parameter inversion, which makes the inversion results for the whole field crop more objective, accurate and universal. The hyperspectral remote sensing image can effectively extract crop growth information from large area and large area.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S127

【参考文献】