摘要
结合Landsat TM遥感影像和环境减灾卫星HJ-1-A、B影像数据,基于面向对象的遥感影像分类技术实现长株潭地区土地利用/覆被的分类提取.综合利用隶属度函数和最邻近分类方法设置分类规则,逐步提取林地、湿地、耕地、人工表面的地物信息.以地形复杂多样的长株潭地区为研究区,收集整理具有代表性的样点用于分类和精度评价.结果表明,利用隶属度函数方法分类结果基本能满足生产者和用户的需要,但是林地、耕地内部二级类精度相对较低,错分比较严重,采用最邻近分类优化分类结果后,研究区总体分类精度达到86.05%,耕地和林地一级类分类精度分别提高到73.63%和87.1%.
This paper tries to extract the land use/cover classification of Changzhutan area by using the object-oriented techniques of muhi-scale image segmentation method from the Landsat TM and HJ-1-A and B. Image segmentation follows the principle of minimum heterogeneity and the segmentation parameters and different characteristic parameters are set to extract different objects. The artificial surface features information of forest, wetland, farmland are extracted gradually according to the classification rules set by comprehensively utilizing the membership function and the nearest neighbor classification methods. Taking Changzhutan area of various and complex terrain as study area, this paper selects the typical and representative sample-spots for classification and accuracy assessment. The results indicate that the classification results by using the membership function can basically meet the needs of producers and consumers, however, the accuracy of the classification for level two of forest land and cultivated land is relatively low, containing many errors; After the k-nearest neighbor classification was used to optimize the classification results, the overall classification accuracy of the study area increases to 86.05%, the classification accuracy of level one of cultivated land and forest land increase to 73.63 % and 87.1 % respectively.
出处
《华中师范大学学报（自然科学版）》
CAS
北大核心
2014年第6期910-916,共7页
Journal of Central China Normal University：Natural Sciences
基金
国家科技支撑计划项目(2009BAI78B03)
国家重大专项-中国科学院战略性先导科技专项(XDA0505107)
国家生态十年计划项目(XDA050501)
关键词
面向对象
多尺度分割
分类规则
土地覆盖分类
object-oriented
multi-scale segmentation
classification rule
land Cover classification
