地球信息科学学报

地球信息科学学报 ›› 2011, Vol. 13 ›› Issue (2): 252-259.doi: 10.3724/SP.J.1047.2011.00252

• 遥感技术与应用 • 上一篇    下一篇

TM图像杨树林识别的MLC与LSU算法应用分析——以河北省文安县为例

赵宇鸾1,2, 李秀彬1, 辛良杰1, 张英1,2   

  1. 1. 中国科学院地理科学与资源研究所,北京 100101;
    2. 中国科学院研究生院,北京 100049
  • 收稿日期:2010-09-25 修回日期:2011-03-17 出版日期:2011-04-25 发布日期:2011-04-25
  • 通讯作者: 李秀彬(1962-),男,研究员,主要从事土地利用变化研究。E-mail:lixb@igsnrr.ac.cn E-mail:lixb@igsnrr.ac.cn
  • 基金资助:

    国家自然科学基金项目(40971062)。

Image Recognition and Extraction of Poplar Plantation Based on MLC and LSU:Case of Wen-an County of Hebei Province

ZHAO Yuluan1,2, LI Xiubin1, XIN Liangjie1, ZHANG Ying1,2   

  1. 1. Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China;
    2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2010-09-25 Revised:2011-03-17 Online:2011-04-25 Published:2011-04-25

PDF (PC)

赞

摘要： 以Landsat5 TM1、TM2、TM3、TM4 、TM5和TM7等图像数据,经预处理后进行植被指数提取和主成分分析,生成13个波段数据集;并用最优指数法(OIF)选取目视解译波段,运用最大似然法(MLC)和线性光谱分解法(LSU)对华北平原农区河北省文安县2007年5月的杨树林地面积信息作了应用分析。结果表明:(1)TM数据中可见光红光波段、近红外波段和中红外波段,以及冠层植被指数和前三个主成分量在杨树信息提取中具有优势;(2)根据波段间相关系数分组计算最优指数值(OIF),可减少计算量;(3)MLC提取杨树林地面积为11 259.84hm2,占研究区总面积的10.95%。经野外实地验证,生产者精度和用户精度分别为84.07%和93.14%,分类和空间制图效果较好;(4)利用亚像元的LSU分析技术,提取地表破碎地类面积,且提取杨树林地面积为13 701.71hm2,达到研究区面积的13.33%(较前者大幅增加)。这可为杨树林资源快速调查监测和林-粮土地合理利用提供参考。

关键词: MLC, LSU, TM影像, 杨树林, 图像识别

Abstract: China is actively implementing the fast-growing and high-yielding timber base construction program in the past few years. There is an increasing number of poplar woodland in the North China Plain over the past decade, and the North China Plain is one of the major grain-producing areas in China. So the expansion of poplar woodland may influence food security, which attracts more and more attention. The area of poplar woodland is a key to the attention. We used TM band 1, band 2, band 3, band 4, band 5, and band 7 of Landsat 5 as fundamental datum and preprocessed them, then produced a new data set of 13 bands by generating vegetation index and principal component analysis. At last, we chose band combination of visual interpretation by using optimum index factor (OIF) and extracted information of area of poplar woodland by using maximum likelihood classification (MLC) and linear spectral unmixing (LSU). The results showed that: (1) Visible red light, near-infrared, mid-infrared, canopy vegetation index, and the first three principal components were dominant to the information extraction. (2) The 13 bands were separated according to correlation coefficient of the bands and then OIF was computed, which can save work time. (3) The area of extracted pixels of poplar woodland by MLC was 11 259.84ha, which accounted for 10.95% of the research area. Through outdoor authentication, the precision of producer and user was 84.07% and 93.14%, respectively, the outcome of classification and mapping was good. (4) LSU is a sub-pixel technology. So the area of extracted pixels of poplar woodland by LSU was better than that by MLC in the area of surface fragmentation. The extracted area by LSU reached 11 259.84ha, which increased largely in the research area. And it accounted for 13.33% of the research area. In a word, the two ways complemented each other, which can supply a sample for the quickly investigation of poplar woodland and the research on land-use conflict between woodland and arable land.

Key words: MLC, LSU, TM images, poplar plantation, image recognition