利用规则对Worldview-2影像进行地物信息提取研究

Ground Objects Extraction from Worldview-2 Images with Rules

研究针对Worldview-2影像的地物特征,采用一种多层次规则的面向对象地物提取方法,通过建立和执行各类地物的提取规则,从多尺度分割产生的影像对象中提取不同地物。实地验证结果表明,该方法提取地物的总精度为84.2%,Kappa系数为0.791。建筑物、道路、耕地和裸地相混合误提现象较多,主要是由于这4种地物的光谱特征相似导致,应选用更敏感的识别参量或建立更高效的识别规则以提高识别精度。

Based on the features of Worldview-2 remote sensing images, this paper used object-orien- ted method to extract ground objects based on multi-level rules. Through the establishment and imple- ment of extraction rules of all types ground objects, different ground objects were extracted from the multiresolution segmentation images. Real-space verification results show that the overall accuracy of this method was 84. 2%, Kappa coefficient was 0. 791. Due to the similarity of spectral characteris- tics, there were a lot of false extractions regarding buildings, roads, farmlands and bare lands. In or- der to improve the accuracy, more sensitive identification parameters should be used or more effective recognition rules should be established.