SAR Image Change Detection Based On Distributed Parallel Clustering Algorithm

SAR image change detection can be applied in many areas, such as natural environmental monitoring, disaster assessment, military combat effectiveness evaluation and urban planning, making SAR image change detection has become a hot research field of remote sensing. In recent years, with the upgrading of radar imaging capabilities, the data size of SAR image becomes bigger and bigger, and traditional serial change detection algorithms have become increasingly difficult to cope with the growing volume of data of SAR images. In this paper, by integrating kernel fuzzy C-means clustering algorithm(KFCM) with distributed parallel computing system, Hadoop and Spark, we propose two SAR image change detection methods, which can take advantages of the computing power and storage capacity of many nodes in the cluster, and can effectively deal with large-scale SAR image change detection problem and accelerate the speed of change detection.1. We propose a distributed parallel SAR image change detection method based on H-KFCM(Hadoop based KFCM) algorithm. In the process of SAR image change detection, using clustering algorithms to classify difference image into changed class and unchanged class takes amount of computation and time. Hadoop is a distributed parallel computing system, which can divide large-scale data into several blocks stored on its distributed file system HDFS, and compute the data blocks distributed by applying Map Reduce, a distributed parallel computing framework. In the proposed method, we integrate KFCM algorithm with Map Reduce framework by redesign KFCM to meet Map Reduce programming model. Using the Map Reduce based KFCM can significantly accelerate the process of analysis of the difference image. The experimental results show that, the proposed method can effectively detect change information in SAR images, and good speedup is achieved with the increase of the number of nodes in Hadoop cluster.2. A distributed parallel SAR image change detection method based on S-KFCM(Spark based KFCM) is proposed in this paper. Spark is a in-memory distributed computing framework, designed for the field of iterative and interactive analysis. Compared to Hadoop’s Map Reduce framework, Spark can be made several times or even hundred times acceleration performance when running the same job. In Spark, resilientdistributed datasets(RDD) is a distributed memory abstraction that lets programmers perform in-memory computations on large clusters. In the proposed method, we integrate KFCM algorithm with Spark framework by using RDD and operations on RDD to implement the procedure of KFCM. Then we can analyze the difference image of the two original SAR images in parallel way by using the spark based KFCM algorithm. The experimental results show the effectiveness of this method, and can achieve about 17 times the acceleration performance compared to Hadoop based method.