论文标题:内蒙古草地沙化遥感监测图像自动分类方法研究
Study on Grassland Desertification Remote Sensing Monitoring in Inner Mongolia
论文作者 刘文敬
论文导师 卢欣石,论文学位 硕士,论文专业 森林培育
论文单位 北京林业大学,点击次数 70,论文页数 57页File Size5623k
2005-05-15论文网 http://www.lw23.com/lunwen_474924967/ 草地沙化监测;分区;非监督分类;分类重编码;主成分变换-逆变换
grassland desertification monitoring;segmentation unsupervised classification;recode Principal component analysis;(PCA)-Renverse principal;component analysis
草地沙化遥感监测的主要指标包括草地生产力指标、土壤指标和地表覆盖指标等。本文主要研究地表覆盖指标的自动分类方法,在保证分类精度不低于目视解译精度的前提下,提高地表覆盖分类的工作效率。 本文以内蒙古自治区为主要研究区域,采用2004年8月MODIS图像,辅以其他空间地理信息,将多种处理方法融合运用,探讨了内蒙古自治区草地沙化监测的空间分区;对不同分区草地沙化地表覆盖指标进行了自动分类研究;根据地面观测数据、目视解译结果,对地表覆盖自动分类结果进行了精度验证,获得了较好的分类效果。主要研究结论: ●根据区域土壤质地信息、干燥度数据,对内蒙古自治区进行分区,可有效提高地表覆盖自动分类精度; ●对不同分区进行非监督分类研究,根据地面状况确定非监督分类初次分类数、参考2004年目视解译结果确定自动分类重编码(Recode)方式,分类精度达到60～70%; ●非监督分类初次分类数及重编码模式可用于内蒙古草地沙化监测遥感图像自动分类过程。 ●主成分变换一逆变换处理是是一种数据压缩和去相关技术,主要目的就是把原来多波段图像中的有用信息集中到数目尽可能少的新的主成分图像中,并使这些主成分互不相关,从而突出有效信息。本文研究结果表明,主分量变换不适用于MODIS数据的内蒙古草地沙化监测工作:
The indexes of grassland desertification remotesensing monitoring include productivity,soil,surface overlay and so on.The thesis work mainly studied how to find an efficiently autoclassification method of remote-sensing image for classifying surface overlay ,which can improve efficiency of image classification at the same time keeping the classify veracity during grassland desertification monitoring.In this study ,several different treatment rooted in different consideration were used to search after the space subarea and autoclassification method the of grassland desertification remotesensing monitoring in Inner Mongolia,while MODIS remote sensing image with middle resolving power as data resouce and other graphics of special factors as reference ,Inner Mongolia Municipality was fixed as research object.At last,the classification results were verified by groud observation data and the result of visual interpretation,the conclusion of which is desirable.Mainly conclusion:· It can availably improve the precision of surface overlay autoclassification that the whole area was segmented into several subareas according to soil texture and aridity.· Unsupervised classification was explored in each subarea ,and the feasible number of the first classification and the appropriate way of recoding were fixed while the precision can reach 60-70%.· The feasible number of the first classification and the appropriate pattern of recoding can be used in the process of grassland desertification remotesensing monitoring in Inner Mongolia.· (PCA)-Renverse principal component analysis is a kind of technique to decrease correlation and stand out the useful information. The study result shows that this thechnique is unfit for grassland desertification remotesensing monitoring in Inner Mongolia Municipality.
|
