Research On Distributed Processing Technology For Image Data In Space Situational Awareness

In recent years,with the continuous development and expansion of space technology in various countries,space resources have become the most precious strategic resource in the military field.Space situational awareness capabilities are the core competencies of ensuring space security,space resource development,and national security.As an important data type in the space situational awareness task,the space situational awareness image data is an ever-increasing data set of the space-borne imaging equipment onboard the satellite.With the increase in the number of exploration satellites and the increase in space activities,this data set will also become increasingly fine-grained in its representation of the space situation,and its growth rate will also be faster and faster.In the face of massive image data in space situational awareness,more studies have focused on the stand-alone processing algorithms for typical images in this PB-level data set.For the sake of computing constraints from single computer,a large number of statistical valuable information in space situational awareness image data set has not been fully mined.Meanwhile,the existing space situational awareness image processing system has been difficult to load the calculation of the massive space situational images.Contradictions between massive space situational awareness image data and existing file systems,as well as contradictions between needs for the calculation of the massive space situational awareness image data and capacity constraints of single computer,will become more and more prominent.In view of the above contradictions,based on an in-depth study of the needs of space situational awareness business,this paper proposes and designs a solution to massive space situational awareness image processing based on Hadoop distributed computing framework.The main work and innovations are as follows:First,we combine image processing with Hadoop to solve the problem that Hadoop cannot process space situational awareness image data.The design goal of the native Hadoop platform is to handle massive text data without the ability to process images.Two new data types that can store images are designed for this purpose,and they are combined with OpenCV to provide rich functions for image processing,simplifying image processing.For the different requirements of image processing,the file input format and file output format of the image are designed,and the solution to the storage problem of small files in Hadoop is discussed.Based on the above work,users can process massive images through MapReduce tasks on the Hadoop platform,enabling the Hadoop platform to have the capabilities of processing images.Second,we use Hadoop framework to solve the space target image qualityassessment task in the background of massive data.Due to the particularity of the acquisition environment for space situational awareness image,the image quality of the satellite-borne imaging device may be degraded due to various degrees of interference.The imaging capability of spaceborne imaging equipment in space environment greatly influences the quality of the information acquisition in space situational awareness.The use environment of the satellite-borne imaging equipment determines that we cannot directly detect it.Therefore,we need to judge the quality of the satellite-borne imaging equipment through space situation awareness image.At the same time,the quality of space situational awareness image also greatly influences the space situational awareness task.Based on the above conditions,this paper explores distributed processing for image quality assessment of space situational awareness image which is a typical mission in space situational awareness in the context of massive spatial images.The research goes into deep study of the cascade support vector machine in parallel support vector machine and its distributed parallelization using Hadoop.A new F2 cascade support vector machine model is proposed for the problems existing in cascade support vector Machines.Through theoretical analysis and experiments,it is verified that our proposed F2 cascade support vector machine can maintain stability and accuracy under the premise of shortening training time.F2 cascade support vector machine combined with LBP features was applied to evaluate the quality of images in space situation awareness task,and its feasibility was verified.Third,we use Hadoop framework to solve the space target recognition task in the background of massive data.The space target recognition task is an important task in the field of space situational awareness.Its recognition effect greatly influences the performance of various space situational awareness tasks.In the context of massive space images,this paper conducts a distributed exploration on space targets recognition which is an important task in space situational awareness.To study the characteristics of space target images in space situational awareness,a combined feature method using Hu invariant moments,affine invariant moments,and corner points is proposed.The KNN algorithm widely used by scientific researchers in the space target recognition task is selected and parallelized to solve its memory and run-time bottlenecks in big data.Domestic research on space situational awareness image processing has not been combined with Hadoop.The research results of this paper can be applied to massive space situational awareness image processing,and have a reference value to parallel computing and data mining in image data in space situation awareness task.