基于多尺度融合的高分辨率影像城市用地分类

Urban Land Classification of High Resolution Images Based on Multi-scale Fusion

基于面向对象的信息提取技术,针对高空间分辨率遥感影像进行城市用地分类。首先针对不同城市地物,选择适宜的提取尺度;然后探讨不同城市地物类型提取的适宜特征,充分利用光谱、空间结构、上下文关系、纹理等信息描述地物;最后融合不同地物多尺度下的提取结果。以北京市部分地区QuickBird影像为例,实现城市用地类型的自动分类,结果表明:该方法应用于城市用地分类的精度高达86.74%,为高空间分辨率遥感影像城市用地分类研究提供了新思路。

With object-oriented techniques of information extraction, high spatial resolution images are used to classify urban land. Different urban land-cove types have different characters, thus their appropriate analysis scale and image features vary greatly for exacting land-cover exactly. During the classification process, first of all, proper extraction scale is chosen for each type of urban land-cover. After that, appropriate image features, such as spectrum, spatial structure, context,texture etc. , are se- lected to describe different types of urban land-cover. Finally, extraction results of all urban land types at different scales are fused together. QuickBird image of part of Beijing is taken as an example and the experimental results show that the classification precision reaches up to 86. 74%. Therefore, the presented method is appropriate to classify urban land with high spatial resolution images.