Research On Distributed Unsupervised SAR Image Change Detection Based On Hadoop

SAR(Synthetic Aperture Radar)image is a kind of the remote sensing image.As one of the main applications of SAR image,SAR image change detection has been widely used in agriculture,environmental monitoring,damage assessment and forest monitoring.In recent years,with the improvement of SAR imaging ability,the amount and scale of SAR image are becoming larger and larger.The traditional SAR image change detection method is difficult to deal with the large number and large scale of SAR image.Hadoop is a distributed parallel computing framework,which can make full use of the computing power and storage capacity of multiple nodes of the cluster to handle big data.In this paper,several unsupervised SAR image change detection methods based on Hadoop are proposed to solve the challenges that traditional change detection methods are facing.1.This thesis proposes two kinds of FLICM(Fuzzy Local Information C-Means)algorithm based on Hadoop.The first Map-FLICM clustering algorithm is proposed for the analysis on difference image of the large number of SAR images.In the Map-FLICM clustering algorithm,the Map of MapReduce is used only,and the FLICM clustering algorithm is used to analyze the difference map in the Map function.Experimental results show that this method achieves 31.45 times speedup compared with serial method.Secondly,a MapReduce-FLICM clustering algorithm is proposed to analyze the difference image in the large scale SAR image change detection.In the MapReduce-FLICM clustering algorithm,the updating of membership degree is calculated in the Map phase of the MapReduce,and the clustering centers are updated in the Reduce phase.Multiple MapReduce jobs achieve the iterative process of FLICM algorithm.When MapReduce-FLICM clustering algorithm is used to analyze the difference map,the SAR image is divided according to the grid,and the overlap area is added at the boundary to keep the neighborhood informations of pixels.Experimental results show that the proposed method can deal with the large scale SAR image effectively.2.In this thesis,we propose a framework by combining Hadoop with OpenCL,which uses OpenCL to transfer computing tasks to heterogeneous devices,improves the efficiency of the analysis on the difference image.In the framework,The MapReduce-KFCM and MapKFCM are implemented for analyze the difference image.The MapReduce-KFCM algorithm requires iterative MapReduce task.The iterative convergence algorithm of machine learning can tolerate the errors caused by the small number of parameters not updated in an iteration,the error will be corrected in the subsequent iteration,and the final algorithm can also achieve convergence.In MapReduce-KFCM,the efficiency of MapReduce-KFCM algorithm is improved by increasing the number of membership updates within a MapReduce to reduce the number of tasks in MapReduce.Finally,the experiments are carried out to verify the correctness and performance of the SAR image change detection method based on MapReduce-KFCM and Map-KFCM clustering algorithm in framework of Hadoop with OpenCL.The experimental results show that compared with the ordinary Hadoop,the maximum speed of difference image analysis is 3.56 times faster.