基于对象级的高分辨率遥感影像分类研究

Classification of High-resolution Remote Sensing Images Using Object-oriented Method

依据高分辨率遥感影像的特点,结合深圳市QUICKBIRD数据提出一种基于多尺度分割的对象级遥感分类方法。文中首先利用分形网络演化法(FNEA)进行多尺度图像分割,获取对地表实体更具代表性的图像对象,然后利用对象所包含的光谱、空间特征来确定地物识别中可能要用到的各种特征参数,最后通过构建语义结构实现了研究区地物的逐级分层分类。研究结果表明,本文所采取的方法比传统方法在分类精度上有了明显的提高,为高分辨率遥感影像的信息提取提供了新的技术途径。

High-resolution remote sensing images have many more spatial characteristics than low-resolution data except spectral characteristics. Object-oriented image classification is a new technique in this research field, it can make most use of their advantages to extract information compared to the conventional pixel-oriented methods. In this case study, we classified QUICKBIRD image of Shenzhen city with the new method. Firstly, the image was multi-scale segmented by Fractal Net Evolution Approach to get objects; and then, we selected some characteristic parameters for realization according to spectral and spatial features of image objects. These different objects could be recognized easily using some suitable characteristics; finally, multiple level classification was realized based on semantic structure in the study area. The result showed that classification accuracy was improved by using object-oriented method, and this approach provided a new way for classification of high-resolution remote sensing data.