Patch-based land cover change detection using multitemporal and multivariate alteration detection of optical satellite imagery

Land-use/land-cover change detection is one of the most important applications of remote-sensing images. With the development of satellite and sensor techniques, the complexity of change detection caused by the improvement of image spatial solution is increasing. The spatial unit of data analysis is generally altered from a pixel to a patch that contains not only spectral information but also local spatial and texture information. For change detection algorithms, previous studies focused on subtracting pixels/objects using a statistical approach called Multivariate Alteration Detection (MAD) from bitemporal images. However, the problem of inconsistency caused by dealing with more than two optical satellite images has not been addressed. This paper introduces a novel method, called Multitemporal Multivariate Alteration Detection (MMAD), to alleviate this problem. This method is based on weighted generalized canonical correlation analysis, which solves canonical coefficients for multivariable and multitemporal data, thereby resulting in consistent change detection. In addition, a new weighting scheme based on object similarity, image quality, and temporal coherence is introduced into MMAD to reduce the sensitivity to large landcover changes and to stably distinguish changed from non-changed area. Moreover, combining patch information with MMAD can improves the detection accuracy and make the proposed method feasible and obtain reasonable results.