南京林业大学学报(自然科学版) ›› 2016, Vol. 40 ›› Issue (02): 167-172.doi: 10.3969/j.issn.1000-2006.2016.02.028

• 研究论文 • 上一篇    下一篇

基于改进分割算法的退耕地树冠信息提取

吴 见1,2,王帅帅1,谭 靖3   

  1. 1.滁州学院地理信息与旅游学院,安徽 滁州 239000;
    2. 安徽省地理信息集成应用协同创新中心, 安徽 滁州 239000;
    3.北京航天泰坦科技股份有限公司,北京 100083
  • 出版日期:2016-04-18 发布日期:2016-04-18
  • 基金资助:
    收稿日期:2015-04-01 修回日期:2015-08-25
    基金项目:滁州学院科研项目(2014PY07); 安徽省高等学校自然科学研究重点项目(KJ2015A265); 北京市科技新星计划(Z131101000413086)
    第一作者:吴见(xiangfeidewujian@126.com),讲师,博士。
    引文格式:吴见,王帅帅,谭靖. 基于改进分割算法的退耕地树冠信息提取[J]. 南京林业大学学报(自然科学版),2016,40(2):167-172.

Tree-crown information extraction in returning farmland to forest land based on improved segmentation algorithm

WU Jian1,2, WANG Shuaishuai1, TAN Jing3   

  1. 1. Geography Information and Tourism College, Chuzhou University, Chuzhou 239000, China;
    2. Anhui Center for Collaborative Innovation in Geographical Information Integration and Application, Chuzhou 239000, China;
    3. Beijing Oriental Titan Tec
  • Online:2016-04-18 Published:2016-04-18

摘要: 优化特征空间和改进分割算法是基于面向对象技术实现退耕地树冠信息准确提取的重要环节,同时也是高分影像地物识别研究中需迫切解决的问题。为此,以北京张山营镇部分区域的QuickBird影像为数据源,根据光谱阈值实现一级分割得到林地区域,同时采用区域进化的区域增长算法对改进均值滤波算法去噪声处理后的全色波段执行二级分割,最后结合形状、光谱、纹理指标构建的特征空间完成了树冠信息提取。结果表明:改进分割算法总体精度达91.5%,Kappa系数为0.836 2,分别较传统方法提高了15.5%和0.120 4。

Abstract: Optimization of feature space and improvement of segmentation algorithm are the important links of accurate tree-crown information extraction in returning farmland to forest land based on the object-oriented technology. They are also an urgent task to solve the problem on the research of feature recognition by high resolution image. QuickBird image of parts of Zhangshanying in Beijing was selected as data source in this study. Firstly, forest area was obtained by the first level segmentation according to spectral threshold. Secondly, the second level segmentation was performed on the panchromatic band dealt with the noise by improved median filter algorithm according to the regional growing algorithm based on regional evolution. Finally, tree-crown information extraction was completed using the feature space built by shape, spectral and texture index. Results showed that the overall accuracy of the method in this paper was 91.5% and the KAPPA coefficient was 0.836 2, increased by 15.5% and 0.120 4 than that of the traditional method respectively.

中图分类号: