Multi‐scale information extraction from high resolution remote sensing imagery and region partition methods based on GMRF–SVM

Abstract

This paper proposes the work flow of multi‐scale information extraction from high resolution remote sensing images based on features: rough classification – parcel unit extraction (subtle segmentation) – expression of features – intelligent illation – information extraction or target recognition. This paper then analyses its theoretical and practical significance for information extraction from enormous amounts of data on a large scale. Based on the spectrum and texture of images, this paper presents a region partition method for high resolution remote sensing images based on Gaussian Markov Random Field (GMRF)–Support Vector Machine (SVM), that is the image classification based on GMRF–SVM. This method integrates the advantages of GMRF‐based texture classification and SVM‐based pattern recognition with small samples and makes it convenient to utilize a priori knowledge. Finally, the paper reports tests on Ikonos images. The experimental results show that the method used here is superior to GMRF‐based segmentation in terms of both the time expenditure and processing effect. In addition, it is actually meaningful for the stage of information extraction and target recognition.

Acknowledgement

This research was supported by National Natural Science Foundation of China (40301030) & (40401039) and Innovation Project of IGSNRR(CXIOG‐D00‐06). Special thanks to Prof. Axing Zhu and Dr Jinyang Du for their reviews and improvement to the English language of the paper. The authors appreciate the anonymous reviewers for their helpful comments.