地理科学进展 ›› 2016, Vol. 35 ›› Issue (2): 255-263.doi: 10.18306/dlkxjz.2016.02.012

• 研究论文| 土地利用 • 上一篇    

基于GF-1与Landsat-8影像的土地覆盖分类比较

宋军伟(), 张友静*(), 李鑫川, 杨文治   

  1. 河海大学地球科学与工程学院,南京 210098
  • 收稿日期:2015-08-01 接受日期:2015-11-01 出版日期:2016-02-10 发布日期:2016-02-10
  • 通讯作者: 张友静 E-mail:695276282@qq.com;zhangyj@hhu.edu.cn
  • 作者简介:

    作者简介:宋军伟(1991-),男,陕西咸阳人,硕士研究生,主要从事遥感技术机理与应用研究,E-mail:695276282@qq.com

  • 基金资助:
    国家科技重大专项(08-Y30B07-9001-13/15)

Comparison between GF-1 and Landsat-8 images in land cover classification

Junwei SONG(), Youjing ZHANG*(), Xinchuan LI, Wenzhi YANG   

  1. School of Earth Sciences and Engineering, Hohai University, Nanjing 210098, China
  • Received:2015-08-01 Accepted:2015-11-01 Online:2016-02-10 Published:2016-02-10
  • Contact: Youjing ZHANG E-mail:695276282@qq.com;zhangyj@hhu.edu.cn
  • Supported by:
    National Science and Technology Major Project of China, No.08-Y30B07-9001-13/15

摘要:

高分一号(GF-1)卫星具有多种分辨率与大幅宽结合、重访周期短等优势,而Landsat-8卫星具有多波段、高辐射分辨率等优势。针对不同传感器参数特点,利用支持向量机分类器(SVM)对同区域同期两种数据进行土地覆盖分类对比研究。结果表明:两种传感器对应波段决定系数均大于0.92;典型样本的光谱趋势一致性良好,但在农田与林地、不透水面与裸土的典型样本可分离性方面,Landsat-8优于GF-1;GF-1与Landsat-8的分类总精度分别为90.38%和90.07%,但不同地物类型的分类精度存在差异,波谱响应函数的差异可能是导致Landsat-8对林地的分类精度高于GF-1的原因;此外,GF-1对零碎分布地物类型的分类精度高于Landsat-8,主要原因是GF-1具有更高的空间分辨率。

关键词: 土地覆盖分类比较, 地物波谱, 地物可分性, GF-1影像, Landsat-8影像

Abstract:

The GF-1 satellite has advantages such as short revisit period and the combination of various spatial resolutions and large swath, while the Landsat-8 satellite has the advantages of multi-channels and high radiometric resolution. This article presents a case study focused on land cover classification in Zhongxiang City, Hubei Province. Considering the characteristics of different sensor parameters, the Support Vector Machine classifier (SVM) was applied to the two datasets of the same region on 6 August 2013 and a comparison was made on the classification results. The result shows that the coefficient of determination of the corresponding channels between these two sensors are over 0.92. Good consistencies are found in typical samples’ spectrum. However, compared to GF-1, Landsat-8 has better separability between farmland and woodland, and between impervious surface and bare soil. The overall classification accuracy of GF-1 and Landsat-8 reaches 90.38% and 90.07% respectively, whereas there are differences in classification accuracies of different surface types. The differences in spectral response functions may account for the advantage of Landsat-8 on woodland identification accuracy as compared to GF-1. In addition, compared to Landsat-8, GF-1 outperforms in classification accuracy on surface types with fragmented distribution because GF-1 has higher spatial resolution.

Key words: comparison of land cover classifications, spectrum of surface types, separability of surface types, GF-1 image, Landsat-8 image