Research On Programming Support Tool For Fast Processing Of Remote Sensing Data

Advances in sensor and computer technology are gradually improving the way remotely sensed data is collected,managed and analyzed.The resolution of the sensor is increasing in the spectral dimension and spatial dimension,and the remote sensing data are becoming more and more characterized by the large scale data volume.With the increase of the complexity of geospatial analysis and modeling,the calculation of remote sensing image processing and information extraction is becoming more and more characteristic of intensive computing.In addition,emergency remote sensing applications often have a high processing time requirements.In addition,emergency remote sensing applications usually have high requirements for real-time processing.With the massive data,complex calculation and real-time requirements,the extraction of remote sensing information has been challenged.Parallel computing is the effective approaches for solving the above problems.However,it is complicate and difficult to implement the parallel implementation of remote sensing algorithm by using low-level parallel programming language.While coping with massive multi-dimensional remote sensing image data and complex remote sensing application algorithms,remote sensing experts must have sufficient knowledge of parallel system architecture.Quickly process programming tools for remote sensing data can simplify the parallel computing and improve the code reusability.The quickly programing tools for parallel remote data processing are required to simplify parallel computing and improve code reusability.In addition,the remote sensing data processing mode is diverse and complex.The parallel computing of numerical calculation and DAG remote sensing data processing algorithms is difficult to implement.The application of parallel computing in remote sensing data processing has been developed for many years.There are the following shortcomings: First,the remote sensing data processing parallel solution is over coupling with the specific algorithm,and the code reusability is poor.Second,remote sensing data has many characteristics of multi-band and metadata,and it is difficult to map multi-dimensional remote sensing data to general-purpose parallel programming skeleton.Third,there are certain risk of stability and reliability in most current remote sensing data parallel processing fast programming tools,which are achieve directly based on low-level programming language.Fourth,the current sensing parallel programming templates usually only provide simple data parallel mode,generally only including point operations,neighborhood computing and global computing.The fast parallel tool for numerical calculation and DAG algorithm are lacking.Based on the analysis of remote sensing data and processing flow,this paper discusses the key issues in the remote sensing data parallel processing tool,then constructs and implements various typical remote sensing parallel algorithm skeleton.The main contributions and innovations of this paper are mainly reflected in the following aspects:(1)The idea of constructing the skeleton of the remote sensing based on the general parallel algorithm skeleton is proposed,and the typical programming templates of remote sensing data are constructed and realized from three progressive levels.The instability risk of using the low-level programming language is avoided.(2)The abstract model of remote sensing data is proposed,and the remote sensing data template is constructed in the form of generic programming.The distributed remote sensing data template is constructed according to the specific rules.The single-band and multi-band remote sensing data are mapped into the Muesli distributed matrix space.(3)The common parallel processing flow is abstracted for the typical remote sensing image processing and information product production algorithms. Based on the Muesli task parallel skeleton,the parallel programming template RSParallel for the remote sensing data is constructed,which provides an exemplary idea for the development of parallel model in remote sensing.(4)The higher level abstraction of the common numerical calculation problem in remote sensing application is proposed.A higher level of the algorithm skeleton RSPallel NLE for solving nonlinear equations is constructing by the fusion of remote sensing data parallel programming template and numerical computing library.(5)Based on the many-task computing,a DAG parallel programming template for remote sensing data processing is constructed.The task representation model is defined based on many-task computing.A dynamic DAG scheduling strategy based on the depth and the critical path is proposed.The above parallel programming templates are tested via several typical remote sensing application algorithms and effective parallel performance is achieved.The theory and experiment show that it is an effective way to abstract the common mode of remote sensing data processing and construct the corresponding parallel programming template.