Research On Classification Acceleration Of Remote Sensing Image In Spark

The primary condition for the wide application of remote sensing technology in various fields is the rapid classification and judgement of remote sensing images.In recent years,with the rapidly developing of remote sensing technology,the resolution of remote sensing data is higher and higher which provides conditions for improving the accuracy of remote sensing data classification.The use of remote sensing images to classify and identify high accuracy of objects has been a research hot spot in the field of remote sensing.The key to accurate classification of remote sensing images is the design and selection of classification algorithms.The shortcomings of traditional machine learning algorithm can not satisfy the accurate analysis of remote sensing data,which hinders the wide application of remote sensing technology.In recent years,deep learning has made great achievements in many fields,and has overcome many problems in artificial intelligence.The advantage of deep learning is that it is good at dealing with highly complex multidimensional data.At present,due to the rapid growth of remote sensing data,the rapid processing of massive remote sensing data under a single machine and the rapid and efficient processing of remote sensing images are currently hot topics in this field.Nowadays,due to the rapid development of distributed platforms,this paper uses distributed clusters to speed up the processing of massive amounts of remote sensing data.This paper studies the accurate and rapid classification of ground objects based on remote sensing images from three aspects of high efficiency extraction of remote sensing data features,depth learning and distributed algorithms.The main research contents are divided into the following three aspects:Acceleration research on feature extraction from remote sensing data under Spark platform.In order to improve the speed of extracting remote sensing data features,and to explore the processing method of extracting remote sensing data features under distributed memory computing environment,this paper uses Spark computing platform to develop accelerated extraction of remote sensing data featurevalues.Considering that the storage way of remote sensing data and the segmentation of remote sensing data will affect the speed of Spark extracting the eigenvalues of remote sensing data.The design experiment is as follows: Firstly,the time of extracting the eigenvalues of remote sensing data is compared with Spark in single machine mode and Spark-standalone mode;secondly,the speed of extracting features of remote sensing data by Spark is compared under the Spark-standalone and HDFS modes;finally,the effect of remote sensing data on the speed of remote sensing data extraction by Spark is divided.Taking Landsat8 as experimental data and extracting the normalized values of vegetation index and other characteristic values as an example,it is known from experiments that:(1)The speed of using Spark to extract the feature data of remote sensing data is about 2 times higher than that of stand-alone mode.The Speed of using Spark to extract feature data from remote sensing data based on HDFS is about 1.2 times faster than that of stand-alone mode.The speed of extracting the eigenvalues of remote sensing data without grid segmentation,Spark extracts the eigenvalues of remote sensing data after grid segmentation by approximately 1.5 times.Compared with the speed of extracting the eigenvalues of remote sensing data without grid segmentation,Spark extracts the eigenvalues of remote sensing data after grid segmentation increase by 1.5 times.(2)The research of grassland discrimination based on Depth Blief Network.In order to improve the accuracy of grassland recognition based on remote sensing data,this paper proposes a deep learning model based on Deep Belief Net to extract grassland.The feature index of Ladsat8 satellite remote sensing data was extracted as the data source of this study.The measured data verify that: the accuracy of the DBN model to grassland recognition is 97.41% by selecting and setting the hidden layers of different features and compared with support vector machine model and decision tree model,the accuracy of grassland remote sensing data was increased by 2.41% and2.21% respectively.(3)Research on parallel acceleration of distributed DBN under Spark.The Distributed Deep Belief Network(DDBN)has the problems of data skew,lack offine-grained data replacement,and cannot automatically cache data with high reutilization,which leads to the shortcomings of the high computing complexity and low operational timeliness of the DDBN Deep.In order to improve the timeliness of DDBN,a parallel acceleration strategy for DDBN data under Spark is proposed,which includes two algorithms of based on Label Set based on Range Partition(LSRP)and Cache Replacement based on Weight(CRW).The problem of data skew is solved by LSRP algorithm,and the problem of Resilient Distributed Datasets(RDD)reuse and insufficient memory space due to excessive cache data.is solved by CRW algorithm.The results show that the training speed of DDBN is 2.3 times higher than that of DBN training under single machine and the distributed parallelism of DDBN is greatly improved by LSRP and CRW.