基于像元级SAR图像时间序列相似性分析的水体提取  

Extraction of water areas based on similarity analysis using pixel-level SAR image time series

作　　者：王亚飞[1] 程亮[1] 李满春[1] 陈伟[1] 陈小雨[1] 谌颂 

机构地区：[1]南京大学江苏省地理信息技术重点实验室,南京210093

出　　处：《国土资源遥感》2014年第3期67-73,共7页Remote Sensing for Land & Resources

基　　金：国家科技支撑计划课题"多尺度地理国情要素监测技术"(编号:2012BAH28B02);国家自然科学基金资助项目"‘空-车’LiDAR点云数据一体化的高质量自动集成方法研究"(编号:41371017)共同资助

摘　　要：针对SAR图像数据集的特点,提出了一种基于像元级图像时间序列相似性的水体提取方法。其基本思想是:构建像元级SAR图像时间序列,选取动态时间归整(dynamic time warping,DTW)算法作为时间序列相似度的度量方法,计算所有像元与标准水体像元时间序列的相似性;将水体边缘混合像元的DTW距离值设定为参考阈值,采用阈值法提取相似性较高的时间序列数据,其对应的像元即被识别为水体像元;最后利用每个像元的DTW距离值代替其后向散射值,采用8邻域搜索方法提高水体识别的精度。以2008年1—12月获取的25景分辨率为150 m的ENVISAT ASAR图像进行水体像元提取试验,结果表明,该方法的完整率和正确率均较高,能够应用于大范围区域水体的提取与制图。In view of the characteristics of SAR images stack data, this paper presents a method for extraction of the water area based on similarity analysis using pixel -level SAR image time series. The basic idea is as follows：Firstly, the pixel-level SAR image time series is constructed, and the dynamic time warping（ DTW） is chosen as the similarity measurement to compute the similarities between all the pixels and sampled water pixel. Then, the water pixels are extracted with the threshold value method, referring to the DTW distance of mixed pixels on borders of water for threshold value. Lastly, by replacing the backward scattering coefficients with DTW distance of each pixel, the accuracy of water identification is improved by 8-connected method. 25 wide-swath ENVISAT ASAR images collected during the period from January to December in 2008, which have 150 m spatial resolution,were utilized to extract water pixels in the study area. The result shows that the water pixels extracted by this method have high accuracy and integrity,thus proving that this method is practical in the extraction and mapping of water body distribution at the large regional scale.

关 键 词：像元级 时间序列 相似性 动态时间归整( DTW) 水体分布提取 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]