Information extracted from high resolution satellite images, such as roads, buildings, water and vegetation, has a wide range of applications in disaster assessment and environmental monitoring. At present, object oriented supervised learning is usually used in the objects identification from the high spatial resolution satellite images. In classical ways, we have to label some regions of interests from every image to be classified at first, which is labor intensive. In this paper, we build a feature base for information extraction in order to reduce the labeling efforts. The features stored are regulated and labeled. The labeled samples for a new coming image can be selected from the feature base. And the experiments are taken on GF-1 and ZY-3 images. The results show the feasibility of the feature base for image interpretation.

Date Published: 18 November 2014
PDF: 8 pages
Proc. SPIE 9263, Multispectral, Hyperspectral, and Ultraspectral Remote Sensing Technology, Techniques and Applications V, 92630B (18 November 2014); doi: 10.1117/12.2068789

Published in SPIE Proceedings Vol. 9263:
Multispectral, Hyperspectral, and Ultraspectral Remote Sensing Technology, Techniques and Applications V
Allen M. Larar; Makoto Suzuki; Jianyu Wang, Editor(s)