面向水土保持监管的黄土高原区生产建设项目地块快速提取

生产建设项目是黄土高原区人为水土流失的主要来源之一.快速获取生产建设项目地块的位置和属性信息,对提升水土保持监管效率、保护区域水土资源具有重要意义.当前生产建设项目地块的提取主要采用专家目视解译遥感影像的方法,存在效率低、主观性强、稳定性差等问题.本文采用高分一号(GF-1)影像,提出一种面向对象的快速提取黄土高原区生产建设项目地块的方法.首先,分析研究区各类典型生产建设项目地块中组成地物的可分性,确定6类用于标记生产建设项目地块的标志性地物.在此基础上,融合多尺度分割和模糊分类方法,构建面向对象的多层次提取模型提取标志性地物,获得生产建设项目地块的提取结果.结果显示:6类标志性地物具备标记黄土高原区生产建设项目地块的能力,使用本方法提取点型生产建设项目地块的数量精度达到95％以上,提取线型生产建设项目地块的长度精度超过86％,且总耗时不超过现行方法的20％.与专家目视解译相比,本文方法在保证提取精度的同时,极大地提升了提取效率,可用于黄土高原较大区域的生产建设项目地块的快速准确提取.这一方法可为水土保持监管的高效化、精准化实施提供有力的技术支撑.

Rapid extraction of land parcels from soil and water conservation supervision-oriented productive and constructive projects in the Loess Plateau

Background]The productive and constructive project (PCP) is one of the main sources of man-made soil and water loss in the Loess Plateau.Acquiring the location and attributes of PCP land parcels is critical to soil and water conservation supervision,management,and protection.Traditional ways of mapping PCP land parcels mainly rely on manual interpretation of remote sensing images,which is time-consuming,subjective and expensive.Methods] Based on GF-1 images,this paper presents a rapid object-oriented method to extract PCP land parcels in the Loess Plateau.Firstly,we analyzed the separability of 29 typical PCP land parcel components in the study area.Six kinds of components were selected as extraction markers.Then,on this basis,an object-oriented multi-level extraction model,which united the multi-resolution segmentation algorithm and the fuzzy classification algorithm,was developed to extract these six kinds of components mentioned above.Final extraction results were achieved after automatic extraction and post-processing procedures.Results] Results showed that six kinds of marked objects,which were buildings and structures,hardened grounds,bare lands,waste slags,coal covered areas and water areas,were reliable to sign PCP land parcels in the Loess Plateau.With the object-oriented multi-level extraction model,the extraction results were considered sustainable and efficient.The quantitative and area accuracy of extracting block-type PCP land parcels was 95.03％ and 85.19％ respectively,while the length accuracy of extracting line-type PCP land parcels was 86.34％.Besides,with the same hardware and software configuration,the overall time was 80％ less than the traditional method.The results validated that the multi-resolution segmentation algorithm portrayed the edges of different objects correctly in several scales,and that the fuzzy classification algorithm described the features of the objects in a more plentiful and accurate way.It was proved that the object-oriented extraction method was suitable for extracting PCP land parcels based on GF-1 imagery.A few of block-type PCP land parcels were failed to be extracted for the broken composition and inadequate extracting area.The interferences of surroundings decreased the extraction accuracy of line-type PCP land parcels.Conclusions] Compared to manual interpretation,this proposed method shows a solid advantage on extraction efficiency with a high accuracy,which can be applied for rapid extraction of PCP land parcels in a large region in the Loess Plateau.This method provides a better technical support for efficient and accurate implementation of soil and water conservation supervision and management.