Parallel Optimization Of High-resolution Remote Sensing Image Detection Based On Cloud Platform

Remote sensing technology relies on various sensors to detect electromagnetic information of surface objects.The acquired high-resolution remote sensing images can be used to describe more detailed spatial features and spectral information,so they are widely used in the field of geoscience research.Anomaly detection and change detection of remote sensing image are the important technologies of remote sensing data analysis,which are of great significance in the fields of resource scheduling,environmental monitoring,urban planning and military target detection.With the continuous improvement of spatial,spectral and temporal resolution of sensors,the scale of remote sensing data also shows a significant growth trend.In the traditional single-machine processing,there will be bottlenecks such as storage and calculation.Therefore,it is urgent to propose technology for remote sensing big data analysis and processing.The cloud computing platform has the big data storage capability and supports distributed parallel computing,so it is a good way to achieve effective processing of remote sensing big data.This paper analyzed the development of remote sensing technology and remote sensing data processing technology,and also studied the related key technologies of cloud computing such as HDFS storage mechanism,MapReduce parallel programming model and Spark distributed framework.On the basis of these studies,distributed parallel anomaly detection and change detection of high resolution remote sensing images based on Spark cloud platform were proposed.In order to further improve the efficiency of remote sensing big data processing,aiming at minimum finish time,this paper designed the task scheduling strategy of cloud computing change detection based on artificial bee colony algorithm.1?This paper designed a distributed parallel optimization method of hyperspectral image anomaly detection,on the basis of studying the hyperspectral image anomaly detection model based on low-rank and sparse representation.Firstly,a distributed parallel k-means algorithm based on map-side pre-aggregation was designed.Then,the background dictionary was constructed in parallel by implement the method of repartitioning.This distributed method computed the model in parallel based on the narrow dependent RDD,which greatly reduced the time consumption caused by the shuffle process.Finally,several sets of comparative experiments were designed.The experimental results show that under the premise of ensuring the detection accuracy,the Spark-based distributed parallel method implements a significant speedup compared with the serial method and has the ability to process large-scale remote sensing data efficiently.2?A distributed parallel optimization for high-resolution remote sensing image change detection was designed.After studying the high-resolution remote sensing image change detection method based on conditional random field model,this paper designed a method for reading multi-temporal images based on multi-RDD association.On the basis of reasonable analysis of algorithm steps and Spark memory computing characteristics,this paper also computed the unary potentials and pairwise potentials in parallel.Besides,RDD cache mechanism was utilized to avoid repeated calculation,which helped improve the effiency of the algorithm processing.The experiment results show that the disributed parallel change detection method achieves high detection accuracy and good acceleration effect.3?This paper proposed a task scheduling method for remote sensing image change detection on cloud platform.Firstly,the distributed optimization task was abstract into DAG's nodes reasonably.Then,a minimize finish time optimization scheduling model with constrainted resource was proposed by considering the start time and running time of the task node.Finally,artificial bee colony algorithm was designed and performed to compute task scheduling model.Experiment results show that effectiveness is improved by reasonable calculation resource allocation and timing control.