A novel change detection method using remotely sensed image time series value and shape based dynamic time warping
Huaqiao Xing, Linye Zhu, Bingyao Chen, Liguo Zhang, Dongyang Hou, Wenbo Fang
Abstract
Satellite image time series change detection methods provide comprehensive understanding of land cover changes. Traditional bi-temporal change detection methods in satellite image time series require consistent time series lengths and use only time series value or shape to calculate change magnitude maps, which may not fully utilize land cover change information. To address this challenge, we propose a change detection method using remotely sensed image time series value and shape based dynamic time warping (TSVS). Change magnitude maps were obtained from the time series trajectories of NDVI and MNDWI using time series value-based dynamic time warping method and time series shape-based dynamic time warping method. Change detection results were derived by clustering the polar coordinate space of time series value and shape using Gaussian mixture model method. Experiments using Landsat images show that the TSVS method improves about 2.75–5.10% compared to the CVA_GMM method, reducing the generation of false alarms.