唐磊 周波 马涛

摘 要：田坎系数测算精度直接影响粮食总产量统计、梯田区土壤侵蚀量测算的准确度，为解决传统的遥感影像目视解译方法提取梯田及田坎精确度不稳定且耗时较长的问题，基于GF-2影像和面向对象的技术方法，确定了各地类最优分割尺度和空间、光谱、纹理等特征参数，建立各土地利用类型提取规则并进行自动提取，其中梯田提取精度为82.55%、Kappa系数达到0.75，田坎面积自动提取精度为68.83%，进而计算的田坎系数为0.153，比实地测量的田坎系数0.151大0.002。阴坡田坎阴影和田坎上林草的投影会导致提取的田坎面积增大，对此可结合后期人工修正，进一步提高精度。

关键词：梯田;田坎系数;GF-2遥感影像;自动提取;提取规则

Abstract：The accuracy of the calculation of the ridge coefficient directly affects the statistics of the total grain output and the accuracy of the calculation of soil erosion in the terraced area. In order to solve the issue that the traditional remote sensing image visual interpretation extraction method for extracting the accuracy of terraces and ridge is not stable enough and takes a long time， based on GF-2 remote sensing image and object-oriented technical methods， the optimal segmentation scale and feature parameters such as space， spectrum， texture， etc. were determined for various types of land and extraction rules for each land use type were established and automatically extracted. Of which， the extraction accuracy of terraces was 82.55%， the Kappa coefficient reached 0.75 and the automatic extraction accuracy of the ridge area was 68.83%. The calculated ridge coefficient was 0.153， which was 0.002 larger than the ridge coefficient that field measured of 0.151. Ridge on shady slopes had shadows. At the same time， the projection surface of the forest and grass growing on the ridge would also increase the area of the extracted ridge. This could be combined with later manual corrections to further improve the accuracy.

Key words： terraces; ridge coefficient; GF-2 remote sensing image; automatic extraction; extraction rules

梯田具有显著的保水、保土、保肥作用，是黄土高原地区重要的水土保持措施[1]。长期以来，基于遥感影像提取梯田信息的方法主要是目视解译法，其依赖解译人员的经验与判读水平，精确度不够稳定且耗时较长[2-3]。随着遥感技术的发展，高分辨率遥感影像的色彩和纹理越来越清晰，为梯田信息的快速、精确提取提供了条件[4-5]。于浩等[6]利用1 m分辨率的IKONOS遥感影像提取陕北地区梯田信息，精度达到81.3%;买凯乐等[7]以分辨率为2.5 m的ALOS遥感影像为数据源，对黄土丘陵沟壑区第Ⅱ副区的燕沟流域进行梯田信息提取，精度达到74.36%;李梦华等[8]利用0.8 m分辨率的GF-2影像数据，将位于黄土丘陵沟壑区第Ⅱ副区的彭阳县划分为4个生态區，采用面向对象的分类方法，分别对4个生态区自动提取梯田信息，精度在65%～77%之间。……