摘要
以中国东北地区完达山以北三江平原(下简称小三江平原)为研究对象,利用2010年多时相HJ-1A卫星和HJ-1B卫星CCD影像作为分类数据源,结合野外调查数据,应用多尺度分割算法,根据影像的丰富信息和物候、时相等特征,采用面向对象的分类方法,进行小三江平原的湿地遥感分类。结果表明,湿地二级分类的总体精度为85.45%,总体Kappa系数为0.81。小三江平原湿地主要分布在低洼地区,总面积为9084.16km2;其中,天然湿地面积为2635.09km2,人工湿地面积为6449.07km2。集中分布的水田是小三江平原最主要的景观类型;天然湿地主要分布在自然保护区、河岸附近和国界周围;其中,草本沼泽面积最大。本研究表明,面向对象的分类方法能够有效利用遥感影像提供的丰富信息,产生较高的分类精度;对于中分辨率遥感影像不同分割尺度的组合使用有利于湿地信息的提取。降雨量较多、较少月份影像的选取和物候规律的利用是提取湿地信息的关键。多时相遥感影像数据的使用,改善了洪泛湿地和森林沼泽的分类效果,尤其是水田边界的划分;同时可以提高湿地分类精度。基于面向对象的遥感方法,利用多时相的中分辨率遥感影像,能够获取较高精度的湿地分布信息,是一种成本较低且行之有效的技术手段。
Remote sensing techniques offer timely,up-to-date,efficient and relatively accurate information for sustainable and effective management of wetland vegetation over a large area.Taking the Sanjiang Plain north of the Wandashan Mountain in Northeast China as an experimental area,quantitative research and studies were made for wetlands classification,with multi-season HJ-1 A,B satellite CCD images as data sources,support of 3S technology and three times field survey data in 2010.Multi-scale segmentation algorithm is applied to extract objects with different segmentation scales.Characteristics such as variety of spatial spectral bands,phenological and seasonal aspect information and the object-oriented classification method were used as well.The results showed that,the overall accuracies of wetland classification are 90.42% and Kappa coefficient is 0.815 2.The overall accuracies of the second level wetland classification are 85.45% and Kappa coefficient is 0.810 8.The object-oriented classification method could efficiently use the image information so that higher classification accuracy could be obtained.The cooperated usages of different segmentation scales are good for wetland information collection in moderate resolution images.Selection of wettest image for a particular time period and understanding of phenological laws are crucial for brief classification of wetland pattern information.The usage of multi-seasonal image data makes the classification of seasonal wetland and floodplain clearer,especially in delineation of paddy fields＇boundaries and classification of floodplain wetlands and forested wetlands,and improves the accuracy of wetland classification.The wetlands were mainly distributed on low-lying areas in study area and the overall area is 9 084.16km 2,with 2 635.09 km2 of natural wetlands and 6 449.07km 2 of artificial wetlands.Paddy field,are the most primary wetland of the study region and the primary landscape type.Herbaceous,marsh follows as the second primary.Natural wetlands mostly locate inside the nature reserve,along the riversides and national boundaries.The rest scatters at other places.It can be concluded that multi-season remote sensing images incorporated with object-oriented processing provide a low-cost and efficient method for meticulous and high-accuracy classification of wetlands in a larger region.
出处
《湿地科学》
CSCD
北大核心
2012年第4期429-438,共10页
Wetland Science
基金
国家重点基础研究发展计划项目(2012CB956103)
中国科学院战略性先导科技专项项目(XDA05050101)
国家水体污染控制与治理科技重大专项项目(2012ZX07207-004)资助
关键词
面向对象方法
多时相HJ-1遥感影像
湿地分类
完达山以北三江平原
object-oriented method
multi-season HJ-1images
classification of wetlands
the Sanjiang Plain north of the Wandashan Mountain
