Object-based Image Analysis of VHR Imagery by Combining Spectrum and Texture

摘要

In this paper, we proposed a contextual and Objectbased Image Analysis (OBIA) classification approach through the experiment, which was using Cognition Network Language (CNL) of commercial software eCognition Developer. Combination of spectrum and texture explored the rich information contents in the Very High-spatial Resolution (VHR) satellite imagery. Meanwhile, contextual relation was adopted in the experiment to improve the classification accuracy. In the research, a few land cover characteristics (contrast, entropy, etc) were studied for the local land covers. By the experiment, six different classes were extracted: building, road, forest, water, farmland and bare soil. 94.05% classification overall accuracy and 0.92 Kappa coefficient were achieved by the approach for the complex land covers. Data set was five bands QuickBird imagery located nearby Kunming Dianchi Lake (KDL), which is a very important highland lake in China, also in the world. The results may be used in lake protection, environment protection, land planning and land use management, and government decision making around KDL.