Distributed SAR image change detection based on Spark

SAR image change detection is a fundamental process in many applications such as damage assessment, natural disasters monitoring and urban planning. Now as the scale of images and the complexity of algorithms rise, sequential methods have been more and more inefficient and powerless. In this paper, we propose a distributed parallel image change detection method based on Spark, an in-memory cluster computing framework, which provides an original support for iterative jobs. The proposed method can make full use of the power of a cluster or a set of commercial computers to process large scale SAR images. Different from the traditional image change detection, a distributed parallel kernel fuzzy c-means clustering algorithm, which is integrated with Spark, is used to part the change map into changed area and unchanged area. Our experimental results on some large scale SAR images show good effectiveness and accelerating performance. Compared to Hadoop based KFCM, the speedup can achieve 18.9 in maximum.