| 摘要: |
| 为对不同农用大棚类型信息进行识别分类和精细化提取，以内蒙古河套灌区不同大棚类型为研究对象，基于Sentinel-2A卫星数据，采用面向对象结合多层多尺度分割技术和阈值分类方法，对大棚类型信息进行提取并对最终提取结果展开精度评价和分析研究。首先利用尺度参数估计（Estimation of Scale Parameter2， ESP2）方法进行了分层分割并优选出最佳分割尺度，在各层最优分割尺度上进行光谱、指数、几何、纹理等特征的提取与优化，获取最优特征组合；然后运用多层多尺度分割阈值分类方法提取不同大棚类型信息。结果表明不同大棚类型信息总体精度达94.8%，kappa系数达0.93。其中：塑料大棚的制图精度和用户精度分别为95.3%和96.6%；单屋面温室大棚制图精度和用户精度分别为88.5%和92.6%。基于多层多尺度分割分类的信息提取方法分别考虑了不同地物最优分割尺度，在不同地物各自的最优分割尺度上提取其信息，以抑制过度分割或亚分割现象，从而降低错分或漏分。因此，高分辨率卫星数据与面向对象多层多尺度分割分类的信息提取方法能够有效提高大棚类型信息提取精度，且能为地物信息精细提取技术体系提供一定参考思路。 |
| 关键词: 大棚类型 精细提取 Sentinel-2A卫星数据 面向对象影像分析 多层多尺度分割分类 |
| DOI：10.11841/j.issn.1007-4333.2024.08.19 |
| 投稿时间：2023-11-10 |
| 基金项目:国家自然科学基金项目（42001366，42261066）；高等学校青年科技人才发展计划项目（NJYT22047）；内蒙古自治区科技计划项目（2021GG0081）；高校基本科研业务费项目-青年教师科研能力提升项目（BR220110） |
|
| Information extraction of agricultural greenhouses types based on object-oriented multi-level multi-scale segmentation |
|
WANG Wuyun1, LI Fei1,2, Hasituya1,2*, Hashengaowa3
|
| (1.College of Grassland， Resources and Environment， Inner Mongolia Agricultural University， Hohhot 10018， China;2.Inner Mongolia Key Laboratory of Soil Quality and Nutrient Resources/Key Laboratory of Agricultural Ecological Security and Green Development at universities of Inner Mongolia Autonomous， Inner Mongolia Agricultural University， Hohhot 010018， China;3.Natural Resources Bureau of Keerqin Zuoqi， Tongliao City， Inner Mongolia Autonomous Region， Tongliao 028001， China) |
| Abstract: |
| In order to refine the extraction and identification and classification of different agricultural shed types， this study takes different shed types in Hetao Irrigation District of Inner Mongolia as the research object， and based on Sentinel-2A satellite data， adopts object-oriented combined with multi-layer and multi-scale segmentation technology and threshold classification method to extract information on shed types and carry out accuracy evaluation and analysis research on the final extraction results. Firstly， the Estimation of Scale Parameter2 （ESP2） method is used to perform hierarchical segmentation and the optimal segmentation scales were obtained for different levels. Then， the spectral features， indices feature， geometric features and the textural features were extracted and optimized on the optimal segmentation scale to obtain the optimal feature combination. A threshold classification method of multi-level multi scale segmentation methodology was developed to map the different types of greenhouses. The results indicated that the overall accuracy of mapping different types of greenhouses achieved 94.8%， and the Kappa coefficient was 0.93. The producer’s and user’s accuracy of plastic greenhouse were 95.3% and 96.6%， respectively. While， the producer’s and user’s accuracy of single roof greenhouse were 88.5% and 92.6%， respectively. The object-oriented multi-level multi-scale segmentation methodology for mapping different types of greenhouses considers the different optimal segmentation scales of different ground objects at different levels respectively， therefore this method can suppress the over-segmentation or sub-segmentation problems and reduce the misclassification/missing ratio. Therefore， the combination of the high-resolution satellite data with object-oriented multi-level multi-scale segmentation method mapping different types of greenhouses can effectively improve the mapping accuracy and provide a certain reference for the technical system for precious mapping of ground object distribution. |
| Key words: greenhouses types precious mapping Sentinel-2 satellite data object-based image analysis multi-level multi-scale segmentation |
