基于QuiCkBird卫星数据的土地利用分类规则集研究

Study on the Rule Set for Land Use Classification Based on QuickBird Image

以天津滨海新区为实验区,研究面向对象技术的高分辨率遥感影像土地利用分类规则集。针对耕地、草地、水域、建设用地、交通运输用地、空闲地的特征差异,综合多尺度的分割特征,尝试不同的分割尺度,最终选定两个最优分割层次,即大尺度层次(分割尺度为400)和小尺度层次(分割尺度为240)。采用有效的特征参数,包括自定义的特征增强参数(NDVI参数、色度放大函数)以及最大差异特征参数、面积参数、不对称性参数、标准差参数,确定各特征的隶属度函数,最终建立分类规则集;应用该分类规则集,通过图层间信息的传递与合并,对实验区的QuickBird(QB)遥感影像进行土地利用分类,精度达86.98%,有效避免了"椒盐现象"。实验证明了面向对象的遥感影像分类方法可充分利用高分辨率影像丰富的信息,有效地提高分类精度。

In this paper,the rule set for object-oriented land use classification based on high-resolution QuickBird image in Tianjin Binhai New Areas is studied.In view of different features of land use types,including arable land,grassland,water area,construction land,transportation land and idle land,two optimal segmentation levels are pointed out after analyzing features of multi-scale segmentation and trying on different segmentation scales.The two segmentation layers are large-scale level with segmentation scale of 400 and small-scale level with segmentation scale of 240.The experiment adopted effective feature parameters,including defined feature-enhanced parameters（NDVI and chromaticity amplification function）,Max.diff.,Area,Asymmetry,and Standard deviation,determined the values of membership functions of different features,and finally established classification rule sets.The object-oriented classification was carried out through adopting this classification rule set and using information transferring and merging between the two segmentation scale levels.The classification accuracy achieved 86.98%,which indicated that the method effectively reduced the ＂Pepper and Salt Phenomenon＂.The experimental results demonstrate that object-oriented land use classification can make full use of the information extracted from high resolution images and thus effectively improve the classification accuracy.