一种新的面向对象城市建筑物信息提取方法研究

Research of Information Extraction of City Building Based on a New Object-Oriented Method

为了提高城市建筑物信息提取精度,改进了常规的面向对象方法,以广州市白云区榕溪街区为研究区,针对其下垫面结构复杂的特性,采用多尺度分割和规则数据库结合的方法自动提取建筑物信息,并通过样本区进行了精度验证,将提取的结果与传统分类方法所得到的结果相互比较.研究结果表明:多尺度分割与构建知识库的方法对于建筑用地信息提取更为有效,与实际值更接近,总体分类精度达到87.01%,Kappa系数为0.87,为面向对象分类方法在城市建筑物信息提取的应用提供了新思路.

To increase the accuracy of information extraction of urban building from high-resolution satellite image, a new object-oriented extraction method based on the general building extraction approaches is proposed, which combines multiple scale segmentation and rule database according to complex urban surface. Baiyun district of Guangzhou city is chosen as the study area, and the building information is extracted utilizing a combination of scale segmentation and rule database. The study results show that the new object-oriented classification method based on rules is more consistent with human way of thinking and close to the actual value. Compared to the tradi- tional pixel-based classification and object-oriented classification, the new classified method can obtain the highest accuracy with overall accuracy of 87.01% and a Kappa of 0. 87, which provides a new application of objected-offented classification to building information extraction.