欢迎访问《应用生态学报》官方网站,今天是 分享到:

应用生态学报 ›› 2017, Vol. 28 ›› Issue (11): 3711-3719.doi: 10.13287/j.1001-9332.201711.012

• • 上一篇    下一篇

基于多源遥感数据的面向对象林分类型识别

毛学刚*, 魏晶昱   

  1. 东北林业大学林学院, 哈尔滨 150040
  • 出版日期:2017-11-18 发布日期:2017-11-18
  • 通讯作者: *mail:maoxuegang@aliyun.com
  • 作者简介:毛学刚, 男, 1981年生, 讲师.主要从事面向对象遥感分类、遥感在林业上应用研究.E-mail:maoxuegang@aliyun.com
  • 基金资助:
    本文由国家自然科学基金项目(31300533) 资助

Object-oriented stand type classification based on the combination of multi-source remote sen-sing data

MAO Xue-gang*, WEI Jing-yu   

  1. School of Forestry, Northeast Forestry University, Harbin 150040, China
  • Online:2017-11-18 Published:2017-11-18
  • Contact: *mail:maoxuegang@aliyun.com
  • Supported by:
    This work was supported by the National Natural Science Foundation of China (31300533)

摘要: 林分类型的识别是森林资源监测的核心问题之一.为研究多源遥感数据协同的面向对象林分类型分类识别,采用Radarsat-2数据和QuickBird遥感影像协同进行面向对象分类.在面向对象分类过程中,采用3种分割方案:单独使用QuickBird遥感影像分割;单独使用Radarsat-2数据分割;Radarsat-2&QuickBird协同分割.3种分割方案均采用10种分割尺度(25~250,步长25),应用修正的欧式距离3指标评价不同分割方案的分割结果,确定最优分割方案及最优分割尺度.在最优分割结果的基础上,基于地形、高度、光谱及共同特征的不同特征组合,应用带有径向基(RBF)核函数的支持向量机(SVM)分类器进行杉木林、马尾松林、阔叶林3种林分类型识别.结果表明:与单独使用一种数据相比,Radarsat-2数据和QuickBird遥感影像协同方案在面向对象林分类型分类方面具有优势.Radarsat-2&QuickBird协同分割方案,以最优尺度参数100进行分割时,分割结果最好.在最优分割结果的基础上,应用两种数据源提取的全部特征进行面向对象林分类型识别的精度最高(总精度为86%,Kappa值为0.86).本研究结果不仅可为多源遥感数据结合进行林分类型识别提供参考和借鉴,而且对于森林资源调查和监测有现实意义.

Abstract: The recognition of forest type is one of the key problems in forest resource monitoring. The Radarsat-2 data and QuickBird remote sensing image were used for object-based classification to study the object-based forest type classification and recognition based on the combination of multi-source remote sensing data. In the process of object-based classification, three segmentation schemes (segmentation with QuickBird remote sensing image only, segmentation with Radarsat-2 data only, segmentation with combination of QuickBird and Radarsat-2) were adopted. For the three segmentation schemes, ten segmentation scale parameters were adopted (25-250, step 25), and modified Euclidean distance 3 index was further used to evaluate the segmented results to determine the optimal segmentation scheme and segmentation scale. Based on the optimal segmented result, three forest types of Chinese fir, Masson pine and broad-leaved forest were classified and recognized using Support Vector Machine (SVM) classifier with Radial Basis Foundation (RBF) kernel according to different feature combinations of topography, height, spectrum and common features. The results showed that the combination of Radarsat-2 data and QuickBird remote sensing image had its advantages of object-based forest type classification over using Radarsat-2 data or QuickBird remote sensing image only. The optimal scale parameter for QuickBird&Radarsat-2 segmentation was 100, and at the optimal scale, the accuracy of object-based forest type classification was the highest (OA=86%, Kappa=0.86), when using all features which were extracted from two kinds of data resources. This study could not only provide a reference for forest type recognition using multi-source remote sensing data, but also had a practical significance for forest resource investigation and monitoring.