A hybrid change detection analysis using high-resolution remote sensing image

Abstract

In order to reduce noise and improve the accuracy of the final change results, in this paper, we presented a hybrid change detection method based on combining pixel- and object-schemes, Firstly, the method obtained the orthogonal difference images using the pixel-based iteratively reweighted multivariate alteration detection (IR-MAD) algorithm, additionally in the process of iterative weighting, we applied the regularized scheme to stable the generalized characteristic equation for the multispectral data. Consequently, image segmentation algorithm was used to extract the image objects where the changes occurred. Finally, object-based classification method was applied to determinate the types of changes. In order to validate the effectiveness and feasibility of the proposed approach, a simple case was done by using the Horgos Port local multi-temporal and multispectral high-resolution image data in Xinjiang. Compared to the pixel-level IR-MAD, the experimental results showed that the overall accuracy has been improved, moreover successfully reduced noise and pseudo small changes in the final result.