基于风云卫星中分辨率数据的农业种植区信息提取方法研究

发布时间：2018-03-29 14:43

  本文选题：分类算法　切入点：农作物　出处：《电子科技大学》2015年硕士论文

【摘要】：及时掌握农作物种植分布情况,对于宏观指导农业生产、合理采取农作物管理措施有重要的现实意义。如何提高作物种植区信息提取精度是近年来遥感科学关注的重要课题之一。本文以华北平原为研究区,结合相应作物的物候信息,首先利用国产风云卫星中分辨率数据,基于分层构建决策树的方法来提取裸地和冬小麦种植区分布图,在此基础上利用MERSI NDVI旬产品,建立多作物提取模型,获取不同年份的农作物种植分布情况。主要研究工作如下:(1)基于分层构建决策树的冬小麦种植区提取。首先选取冬小麦生长季中多幅数据质量好的MERSI数据,采用分层提取的方法,对于不同的层次选用与待提取类别最为敏感的特征波段来构建相应的决策树,从中将每一幅影像中冬小麦和裸地提取出来,然后将多幅数据融合为一幅生长季内的冬小麦种植区和裸地分布图,最后采用野外实地调查的数据和LANDSAT 8影像解译数据对提取结果进行精度验证,并对比和分析分层构建决策树与其他分类方法的优劣。提取方法确定后,倒推回2010年,做出不同年份的冬小麦种植分布图。研究表明,采用分层构建决策树获取的精度比其他的分类方法相对较好,其中得到2013-2014年度的裸地、冬小麦和总体提取精度最好,分别达到91.80%,90.19%,90.90%,同时从县域尺度上,MERSI数据提取的冬小麦和裸地与LANDSAT 8影像在空间分布上,大体一致。(2)基于NDVI时序的多作物种植区提取。首先生成不同年份的250米分辨率的MERSI NDVI旬产品,再利用提取出不同年份的冬小麦种植区和裸地对选取的NDVI影像进行掩膜处理,以求取华北平原春玉米、夏玉米和棉花关键生育期内NDVI变化曲线,并结合相应作物的物候信息,建立提取模型,最后利用野外采集的数据和LANDSAT 8影像解译数据对结果进行精度验证,结果表明,不同年份总体提取都达到84%以上,从250米遥感制图及作物遥感监测方面来说,精度能满足需求,同时从县域尺度上,获取的作物种植分布与LANDSAT 8影像大体一致。(3)分类系统设计与实现。在了解本文所涉及的分类算法和研究成果基础上,结合数据的处理框架,从系统应用需求和设计思路上合理开发相应的分类系统,该系统由3个模块组成,包括训练样本数据格式转换模块、分类方法选择模块和数据融合处理模块。
[Abstract]:To grasp the distribution of crop cultivation in a timely manner, and to guide agricultural production at a macro level, It is of great practical significance to rationally take crop management measures. How to improve the precision of information extraction in crop growing areas is one of the important topics of scientific concern in remote sensing in recent years. In this paper, the North China Plain is taken as the research area and the phenological information of the corresponding crops is combined. Based on the resolution data of domestic wind-cloud satellite and the method of constructing decision tree in layers, the distribution map of bare land and winter wheat planting area is extracted. On this basis, the multi-crop extraction model is established by using MERSI NDVI ten-day products. To obtain the distribution of crop planting in different years. The main research work is as follows: 1) extracting winter wheat planting area based on hierarchical decision tree. Firstly, MERSI data with good quality in winter wheat growing season are selected. By using the method of hierarchical extraction, the decision tree is constructed by selecting the most sensitive feature bands for different levels and extracting winter wheat and bare land from each image. Then, several pieces of data were fused into a map of winter wheat growing area and bare land in growing season. Finally, the accuracy of the extracted data was verified by field survey data and LANDSAT 8 image interpretation data. The advantages and disadvantages of hierarchical decision tree construction and other classification methods are compared and analyzed. After the determination of the extraction method, the distribution map of winter wheat planting in different years is drawn back to 2010. The precision obtained by using hierarchical decision tree is better than that of other classification methods. The bare land of 2013-2014 is obtained, and the precision of winter wheat and total extraction is the best. They reached 91.80 and 90.19, respectively. At the same time, winter wheat and bare land and LANDSAT 8 images extracted from the county scale were distributed in space. The extraction of multi-crop planting area is based on NDVI sequence. Firstly, 250m resolution MERSI NDVI products are generated in different years, and then the selected NDVI images are masked by extracting winter wheat planting areas in different years and bare land. In order to obtain the NDVI variation curves of spring corn, summer maize and cotton in North China Plain, and combined with the phenological information of the corresponding crops, the extraction model was established. Finally, the accuracy of the result is verified by the data collected in the field and interpreted by LANDSAT 8 image. The results show that the total extraction in different years is more than 84%, and the precision can meet the demand in terms of 250m remote sensing mapping and crop remote sensing monitoring. At the same time, on the county scale, the obtained crop planting distribution is roughly consistent with the LANDSAT 8 image. The classification system is designed and implemented. On the basis of understanding the classification algorithms and research results involved in this paper, combined with the data processing framework, The system consists of three modules, including training sample data format conversion module, classification method selection module and data fusion processing module.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S512.11;S127
，

本文编号：1681591