Unsupervised SAR Image Change Detection Based On Spark

Synthetic aperture radar(SAR)image change detection has played an important role in many areas,such as natural disaster monitoring,city construction and agricultural statistics.Therefore,SAR image change detection has become one of the hottest research direction in remote sensing field.However,with the development and maturity of sensor technology,remote sensing field has gradually entered the era of big data,and the scale and the quantity of SAR images constantly increase.The traditional serial algorithm has been powerless to deal with the large scale SAR image change detection effectively.In this paper,the fuzzy local information C-mean clustering algorithm(FLICM)and kernel fuzzy C-means clustering algorithm(KFCM)are combined with Spark distributed framework,and then integrated with coprocessors to accelerate.In this way,we puts forward two kinds of SAR image change detection methods based on Spark framework.Spark can make full use of cluster computing and storage ability to effectively deal with the issues of large scale SAR image change detection.In addition,coprocessors can enhance the computing ability of each node and highly improve the efficiency of change detection.1.A SAR image change detection algorithm based on Spark-FLICM algorithm is proposed,which uses the distributed data processing mechanism of Spark.The proposed algorithm can improve the efficiency of change detection and complete the large-scale SAR image change detection quickly.The algorithm sends neighborhood information through the broadcast variable,then call the broadcast data in the closure function.The membership is calculated in Map phase,then clustering centers are computed in Reduce phase through the reduction operation.Under the premise that the accuracy is permitted to change a little,broadcast operation is reduced by increasing the local clustering process,which greatly improves the computational efficiency of the algorithm.The experimental results show that the Spark-FLICM algorithm is correct compared to the serial FLICM algorithm.When there are 4 CPU above in use,the proposed algorithm is more efficient,and it has the scalability with increasing data quantity and number of CPU cores.2.A SAR image change detection method based on coprocessor-accelerated Spark-KFCM algorithm is proposed,which uses Spark framework to transfer the data to every nodes.By integrating JNI technology and Open CL framework,the compute-intensive tasks areoffloaded to the coprocessors for processing,then the results returns to Executor and is packaged as a RDD for subsequent computing.After the parallel design of KFCM algorithm,the process of computing membership and part of computing clustering centers are offloaded to the coprocessors,then the main processor will collect the results obtained from coprocessors and get the clustering centers for the next iteration.The proposed framework can provide higher computing power through the same cluster.The experimental results show that the Spark CL-KFCM algorithm can achieve the same detection accuracy compared with the serial algorithm.Under the circumstances the there are different data quantity and resource,the proposed algorithm has the scalability.Compared to standalone Spark-KFCM algorithm,there is a certain speedup with different CPU cores.