Object-oriented change detection method based on adaptive multi-method combination for remote-sensing images

ABSTRACT

In this study, we propose a novel object-oriented change detection method for remote-sensing images. First, the Gabor texture and Markov random field texture are extracted based on the remote-sensing images, and an initial pixel-level change detection result is produced. Second, in order to reduce the influence of feature uncertainty on the change detection results, the weights of different features are calculated by the Relief algorithm based on the initial pixel-level change detection result, and several difference images are fused to obtain a single comprehensive difference image. Third, different pixel-level change detection results are obtained using diverse change detection methods. The two-temporal images are then stacked and segmented, and to ensure change detection method separability, the weighted object change probability is obtained by fusing five different object change probabilities, which are calculated from the pixel-level change detection results. Finally, the objects are labelled as the class with a higher weighted object change probability. Our experimental results showed that the accuracy of change detection results obtained using the weighted object change probability was higher than that of the change detection results produced using the independent object change probability.

Acknowledgements

This work was supported in part by the National Natural Science Foundation of China (41331175), the Fundamental Research Funds for the Central Universities under Grant (2015XKQY09), and China Postdoctoral Science Foundation funded project.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China [41331175]; Fundamental Research Funds for the Central Universities [2015XKQY09]; and China Postdoctoral Science Foundation.