Research And Design Of Accelerated Processing Algorithms For Satellite-based Astronomical Images

The development of aerospace science makes satellite missions gradually complex,requiring real-time processing of data captured by on-board payloads.In this paper,we take astronomical image subtraction algorithm as an example and develop the research of in-orbit real-time processing of astronomical image algorithm based on CPU-GPU heterogeneous embedded platform.Image difference algorithm is an effective method for searching transient objects or identifying objects with time-varying brightness in time-domain astronomy.Usually,the astronomical image difference method is performed by taking two aligned images of the same observation region,calculating a spatially varying convolution kernel for both images,and using the convolution kernel to obtain the difference image.By systematically observing specific regions of the sky and tracking our sources of interest,a large amount of valuable observational data can be obtained.In modern astronomical research,advanced wide-field telescopes have been introduced in large numbers for long-term and continuous observations,producing a very large number of images and measurements.So that,astronomers are faced with unprecedentedly large data sets.These massive datasets pose new challenges to astronomers in terms of data detection and processing.In order to adapt to the trend of massive data in astronomy,improve the performance of the algorithm,and meet the constraints of in-orbit real-time processing,this paper proposes a frequency domain astronomical image subtraction algorithm based on overlap and save method.According to the time domain convolution theorem,the algorithm transforms the most time-consuming convolution-subtraction operation in the time domain of the algorithm into a less complex multiplication-addition operation in the frequency domain.Also,the algorithm uses the overlap save method to improve the computational efficiency of the algorithm under the condition of spatially transformed convolution kernel by using the overlap preservation technique,and reduces the memory occupation.In addition,for the need of real-time image subtraction algorithm in astronomy projects,this paper analyzes the feasibility of heterogeneous parallel acceleration of the frequency domain astronomical image subtraction algorithm based on overlap save method,according to the characteristics of data independence in astronomical image subtraction algorithm.Based on the above analysis,a detailed parallel design scheme and implementation scheme of the algorithm on CPU-GPU heterogeneous embedded platform is completed.Finally,the whole algorithm is optimized using data parallelism,task parallelism,memory optimization,and other related strategies.The data shows that compared with the original astronomical image difference algorithm,the frequency-domain astronomical image difference algorithm based on overlap and save can not only fit a larger image sub-region,but also can ensure the accuracy of the image after subtraction.In this case,the memory footprint is reduced by 1.26 times,the algorithm complexity is reduced by 1.91 times,and the processing speed of the entire algorithm is increased by 1.213 to 1.46 times.In addition,this paper compares the processing time of the improved astronomical image difference algorithm on the CPU computing platform and on the CPU-GPU heterogeneous computing platform.Compared with the CPU platform,the frequency-domain astronomical image difference algorithm based on CPU-GPU heterogeneous parallelism proposed in this paper increases the processing speed by 1.79 to 2.48 times,which basically satisfies the constraints of the algorithm for real-time processing of on-board devices.In this heterogeneous parallel computing system,the frequency-domain astronomical image difference algorithm based on overlap preservation can not only meet the fast and real-time requirements of deep space exploration missions,but also meet the goals of low memory and low power consumption of on-board computers.The on-orbit adaptive scientific detection of spacecraft provides the possibility.