摘要
由于建筑物的材质不同、高楼阴影等使居民地的自动提取成为遥感技术的一个难点,而且,在干旱、半干旱区,泥质房顶的居民地与泥质荒漠有异物同谱现象使居民地信息提取更加困难。准确了解防护林草的变化信息对维护沙漠铁路、公路的正常运行,保护绿洲都有重要的意义,而防护林草与荒漠草原因为有相同的荒漠植被类型而波谱相似,所以在干旱、半干旱地区土地利用/覆盖的遥感影像分类,普遍受到同物异谱和异物同谱现象的影响和制约,使传统基于光谱特征的分类方法精度低。仔细分析了研究区典型地物的光谱特征和同物异谱与异物同谱现象,并结合实地考察建立了分类体系。基于知识对监督分类的结果做后处理,准确地提取了居民地和防护林草类型,并对分类结果进行了精度评估。结果表明:与传统的基于光谱特征的分类方法相比,基于多特征的遥感影像分类精度明显提高。
In remote sensing technology, it is difficult to extract residential information automatically from image because many buildings are made of different materials and the high-rise has shadow, furthermore, the buildings＇ roof made of mud have the same spectra character with some of the desert, in light of this, it made harder and harder to extract residential information from remote sensing image in arid and semiarid regions. It is great helpful to maintain the operation of desert-railway and desert-road, protect oasis if we clearly knowing the change information of the grass of shelter belt, but the grass of shelter belt have the similar spectra character with desert shrub because of the same kind of vegetation. Therefore the traditional auto-classification based on the spectral character always has poor accuracy because of the common existing of same object having different spectral character and different object having same spectral character especially in arid and semiarid regions. In this paper, we studied the spectral character of our study region in detail and based our classification system on the spectral character and our filed work. After supervised classification we extracted residential area and the grass of shelter belt based on knowledge discovery, finally we got the accuracy assessment reports. Experiment results show that the classification method based on multi-features has great accuracy compared with traditional autoclassification based on the spectral character.
出处
《遥感技术与应用》
CSCD
2006年第2期154-158,T0001,共6页
Remote Sensing Technology and Application
基金
教育部"春晖计划"资助项目(Z2004-1-62006)
关键词
土地利用/覆盖
多特征分类
知识发现
腾格里沙漠
Land use/cover, Multi-features based classification approach, Knowledge discovery, Tengle desert
