论文部分内容阅读
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。
In view of the high spatial resolution and relatively insufficient spectral resolution of ADS40 images, an object-level remote sensing classification method based on multi-scale segmentation is proposed. Firstly, the image object is obtained by multi-scale segmentation, and then the spectral features, geometric features and topological features of the object are used to determine the various feature parameters that may be used in the object recognition and to establish the hierarchical structure of the objects. Finally, Using fuzzy classifier to classify by level to extract feature information. The results show that the object-oriented classification method can significantly improve the classification accuracy and effectively suppress the occurrence of “salt and pepper ” compared with the traditional method, and is more suitable for the automatic recognition classification of ADS40 images with rich geometric information and structural information. The classification of ADS40 images in Taiyuan has verified the effectiveness of this method.