Research And Implementation Of Parallel Processing Technology For Remote Sensing Image On Cloud Architecture Storage And Classification Algorithm

Nowadays,remote sensing images have more and more spectral bands and higher resolution,and are widely used in various fields.However,this situation has led to some problems,such as how to efficiently manage remote sensing data and classify remote sensing images through machine learning.For storage management,the traditional single machine has the disadvantages of small capacity and poor scalability,which cannot meet the storage requirements of remote sensing data.Therefore,researchers use distributed storage mechanisms to classify remote sensing images.However,the current distributed storage mechanism has problems such as slow indexing and waste of storage space,which makes it impossible to efficiently manage remote sensing images.In terms of image classification,due to current It is difficult to obtain remote sensing images,and the resolution of the obtained images is also higher and higher.The traditional method of running machine learning in a single machine environment is less efficient in classifying remote sensing images.In view of the above problems,this paper debugs the traditional WebGIS platform with Hadoop and Spark as the underlying architecture,realizes the parallel processing of remote sensing images,and designs the parallelization of remote sensing image classification based on residual network in Spark computing platform.Method for classifying remote sensing images.The main work is:(1)Hadoop and Spark are used as the underlying architecture to improve the WebGIS platform,so that it can store remote sensing images in HDFS,then use GeoHash and HBase to establish an indexing mechanism to store the data in HDFS twice.Finally,use SpatialRDD to perform remote sensing images.(2)The study uses the residual network to classify remote sensing images in a single machine,and analyzes the classification accuracy and time performance through experiments.(3)The method of parallelization of residual network classification is implemented by TensorFlowOnSpark framework.The method is implemented by using model parallelization and synchronous update model.Finally,compared with the residual network remote sensing image classification results in stand-alone situation,although the classification accuracy is better than that of stand-alone The accuracy is slightly lower,but the time performance is greatly improved,which proves the reliability of the parallel computing framework based on the Spark platform.