学术论文

      基于混合特征空间MRF(Markov Random Filed)模型的高分辨率遥感影像水体提取

      A New Algorithm Based on Hybrid Feature Space MRF(Markov Random Filed)Model for Water Information Extraction from High Resolution Remote Sensing Imagery

      摘要:
      水体信息提取是遥感图像在水资源调查与利用、水生态监测等应用中的关键技术之一.针对现有的水体指数法或影像分类法在水体边界处理效果不够精确、易产生误提取和漏提取等问题,提出一种基于混合特征空间与MRF模型图像分割算法的水体提取新算法.结合遥感图像颜色特征与归一化差异水体指数NDWI创建混合特征空间,将遥感图像中的像素作为MRF模型中的随机变量,构建基于混合特征的能量函数,使用迭代的图割算法(Graph Cut)最小化能量函数确定水体边界,然后根据已提取的水体主体的水体指数及颜色特征等信息对水体边界进行自适应精细化处理.对石梁河水库水体提取的实验表明,该方法能够自动对周边环境复杂的水库水体信息进行提取,并且水体边界的提取效果良好,达到较高的水体提取精度.
      Abstract:
      Water information extraction in remote sensing images is an important application of remote sensing technology in water resources surveying and utilization,detection of water ecology change and other aspects.The existing water extraction methods as water index or image classitication are not accurate enough for water boundary treatment,and they are easy to produce the problem of error extraction and leakage extraction.Based on the existing algorithms that constructing water index to extract water information,we have proposed a new algorithm which combine image segmentation algorithm based on MRF model with normalized difference water index(NDWI) for extraction of water information.We represent the pixels in a remote sensing image as random variables in an MRF model,and introduce hybrid feature in the energy function on these variables,minimize the energy function to find the optimal water boundary,by using an iterative graph cut scheme.Then water boundary is adaptively refined according to the water index and color feature of the extracted water body.The experiment of water information extraction in Shilianghe Reservoir shows that our approach can achieve significant accuracy as it automatically adapts to the extraction of water information in reservoir whose surroundings are complicated and the boundary of water bodies is handled precisely.
      作者: 李士进 王声特
      Author: Li Shijin Wang Shengte
      作者单位: 河海大学计算机与信息学院,江苏南京,210098
      年,卷(期): 2017, 40(1)
      分类号: TP753
      在线出版日期: 2017年6月2日
      基金项目: 国家自然科学基金,江苏省重点研发计划(社会发展)项目