土地利用/覆盖链式遥感分类方法研究

Chain Classification Method Using Remote Sensing Imagery for Land Use/Cover Change Analysis

针对目前大面积土地利用/覆盖变化应用中存在的当地面参考数据不足时分类精度不高和分类过程复杂的问题,本文介绍了一种新型的链式遥感分类方法。该方法利用有限的地面参考数据,根据相邻影像间重叠区域的信息即可完成大面积的土地分类。研究利用该方法对华北地区约1.0×105km2的区域分别进行了林地和非林地的简单分类和耕地、水域、居民地及林地的复杂分类试验,总体分类精度分别为95.0244%和92.0947%,并分别与传统的分类方法进行了对比,总体分类精度仅降低1.6773%和2.1569%。研究结果表明,链式分类精度损失不大,但时间效率大大提高。

For land use/cover change of large area at present,there are problems on low classification accuracy and complex classification processes when ground reference data is not adequate.A new land classification method based on chain remote sensing classification is introduced in this paper.Limited ground reference data is used in the method.Through the information in the overlapping areas of neighboring scenes,classification for land use/cover change of large area is completed.Using this method,simple classification of two classes including forest and non-forest and complex classification of four classes including farmland,water,residence and forest are performed in Northern China with the area of 100,000 square kilometers.The total classification accuracy is 95.0244% and 92.0947%,respectively.Compared with the accuracy using traditional classification method,the total accuracy loss is 1.6773% and 2.1569%,respectively.The result indicates that chain classification has low accuracy loss and high classification efficiency.