中国农业科学 ›› 2010, Vol. 43 ›› Issue (17): 3529-3537 .doi: 10.3864/j.issn.0578-1752.2010.17.006

• 耕作栽培·生理生化·农业信息技术 • 上一篇    下一篇

基于中分辨率影像的农田管理单元自动提取研究

张竞成,顾晓鹤,王纪华,黄文江,何馨,罗菊花

  

  1. (浙江大学环境与资源学院农业遥感与信息技术应用研究所)
  • 收稿日期:2010-01-12 修回日期:2010-02-28 出版日期:2010-09-01 发布日期:2010-09-01
  • 通讯作者: 顾晓鹤

Automatic Extraction of Farmland Management Unit Based on Moderate-Resolution Images

ZHANG Jing-cheng, GU Xiao-he, WANG Ji-hua, HUANG Wen-jiang, HE Xin, LUO Ju-hua
  

  1. (浙江大学环境与资源学院农业遥感与信息技术应用研究所)
  • Received:2010-01-12 Revised:2010-02-28 Online:2010-09-01 Published:2010-09-01
  • Contact: GU Xiao-he

摘要:

【目的】结合“面向对象”的农田管理思想,提出农田管理单元(farmland management unit,FMU)的概念及自动提取方法,并在此基础上探讨和评价中分辨率影像用于农田管理单元自动提取的可行性及效果。【方法】以江苏省2006年一景Landsat 5TM影像中两块典型区域为例,通过决策树分类和多尺度分割等方法实现FMU自动提取。结合人工解译研究区SPOT-5高分辨率影像得到的地块边界信息,对FMU图斑内像元异质性和地块边界吻合度相关指标进行计算和分析。【结果】试验区内作物地块的总体分类精度均超过90%。两块试验区内反映FMU图斑异质性的平均标准差和平均极差分别较全区作物地块整体低70%以上和45%以上;反映FMU地块边界吻合度的误分地块率和面积偏差率均低于10%。此外,多尺度分割中层权重、分割尺度、形状因子和紧凑度因子的设置对FMU的自动提取效果有不同程度的影响。【结论】基于中分辨率影像的FMU自动提取方案基本可行,在研究区内能够获得单元内异质性较低且单元边界与地块边界吻合度较高的提取结果,符合农田管理的要求。

关键词: 农田管理单元, Landsat 5TM, 多尺度分割, 异质性

Abstract:

【Objective】 In this paper, incorporating the idea of “object-oriented” farmland management, the conception of farmland management unit (FMU) as well as its automatic extraction technique were pat forward. The moderate-resolution remotely sensed data were taken as the data source for the first time. Its capability in application was thus evaluated and discussed. 【Method】 Two typical regions in Jiangsu Province were selected within one scene of Landsat 5TM image which acquired in 2006. The FMUs in both regions were extracted through the processes of decision tree classification and multiresolution segementation. With the help of the exact farmland boundaries that digitized from the SPOT-5 high resolution images in both regions, several parameters that corresponding to the heterogeneity of the FMUs as well as the coincidence of boundaries between FMUs and farmland parcels were calculated and analyzed. 【Result】 The total accuracy of classification in both regions was over 90%. The average standard deviation and average extreme difference of FMU which reflect the heterogeneity were lower than corresponding value for entire farmland range at over 70% and 45%, respectively. The misclassified ratio and overlapping degree of FMU which reflect the coincidence of boundaries were lower than 10% for both regions. Besides, the setting of relative parameters that involved in the process of multiresolution segmentation such as the layer weight, segmentation scale, shape factor and compactness factor had a certain impact on FMU extraction. 【Conclusion】 The automatic extracted FMUs can basically satisfy the requirement of a relatively low heterogeneity of the FMU and a high coincidence of boundaries. In addition, to attain a rather ideal extraction result, the user would be better to conduct a configuration of layer weight, segmentation scale, shape factor and compactness factor according to varied planting structures and conditions.

Key words: farmland management unit, Landsat 5TM, multiresolution segmentation, heterogeneity