基于对象级的ADS40遥感影像分类研究

Object-Oriented Approach for ADS40 Image Classification

针对ADS40影像的空间分辨率高而光谱分辨率相对不足的特点,提出了一种基于多尺度分割的对象级遥感分类方法。首先通过多尺度分割获得影像对象,然后利用对象所包含的光谱特征、几何特征、拓扑特征来确定地物识别中可能要用到的各种特征参数,并建立对象间的分类层次结构图,最后利用模糊分类器逐级分层分类来提取地物信息。研究结果表明,面向对象的分类方法与传统方法相比,可显著提高分类精度,有效抑制"椒盐现象"的产生,更加适合于几何信息和结构信息丰富的ADS40影像的自动识别分类。通过对太原市ADS40影像进行分类验证了此方法的有效性。

ADS40 images with high spatial resolution have many more spatial characteristics than low-resolution image except spectral characteristics. We introduced a object-oriented classification method based on multi-scale segmentation to classify ADS40 image of Taiyuan City. Firstly, the whole image is multi-scale segmented to get objects. Then, The features of objects, such as spectral, geometrical and topological characteristics, were measured. The hierarchical structure for classification was built. Finally, we applied a fuzzy rule classifier to extract the classification information of ADS40 images. The research shows that the object- oriented method can improve the overall classification accuracy of ADS40 images, reduce the Pepper and Salt＇ Pheomenon effectively, and meet the requirement of ADS40 images classification compared with classical classification approaches.