海岸带水体遥感信息全自动提取方法

Full-automatic Method for Coastal Water Information Extraction from Remote Sensing Image

针对海岸带水体的特点提出一种通过"像元—对象"及"全域—局部"的双重尺度转换方法,综合利用海岸带水体遥感信息的光谱特征、空间特征、统计特征和不确定性,有机结合地学知识与数据挖掘方法,实现海岸带水体遥感信息的全自动提取。对中国四大海域不同类型海岸带的Landsat ETM+遥感影像进行试验,提取结果具有较好的完整性和连续性,相对精度优于监督分类法和阈值分割法。

A new water-extraction method,which focuses on the characteristics of the coastal water,is proposed.The method adopts scale transformation from pixel to object and from global to local,utilizes spectral features,special features,statistic features and uncertainties of remote sensing information,and makes use of combination of geo-knowledge and data mining technology.Experimental results indicate that,the method can extract water in various types of coastal zones with better completeness,better continuity and higher accuracy than classic methods such as supervised classification and threshold segmentation.