基于跳白块编码和深度神经网络对脉冲星候选体诊断图像的压缩研究  

Research on Compression of Pulsar Candidate Diagnostic Images Based on WBS and Deep Neural Network

作　　者：姜家涛[1,2,3] 谢晓尧 于徐红[1,3] Jiang Jiatao;Xie Xiaoyao;Yu Xuhong(Guizhou Key Laboratory of Information and Computing Science,Guizhou Normal University,Guiyang 550001,China;School of Mathematical Science,Guizhou Normal University,Guiyang 550001,China;FAST Early Science Data Center,Guiyang 550001,China,Email:[e-mail])

机构地区：[1]贵州师范大学贵州省信息与计算科学重点实验室,贵州贵阳550001 [2]贵州师范大学数学科学学院,贵州贵阳550001 [3]FAST早期科学数据中心,贵州贵阳550001

出　　处：《天文研究与技术》2022年第5期470-478,共9页Astronomical Research & Technology

基　　金：国家自然科学基金(U1831131,U1631132,U1731238,11743002);中国科学院天文大科学研究中心FAST重大成果培育项目(FAST [2019sr04]);中国科学院战略性先导科技专项(B类)(XDB23000000);国家重点研发计划(2017YFA0402600);贵州省科技厅联合基金(黔科合LH字[2017] 7338号)资助。

摘　　要：500 m口径球面射电望远镜(Five-hundred-meter Aperture Spherical radio Telescope,FAST)脉冲星搜索产生的候选体诊断图量级呈指数增长,给科学数据管理工作带来挑战,迫切需要研究压缩方法,实现诊断图的有效存储,加快其在网络中传输共享。脉冲星诊断图像由稀疏的黑白图像、随机分布的灰度图和彩色图像组成,简单视为彩色图像用同一种压缩方法处理显然不合理。提出跳白块编码和深度网络压缩编码压缩模型对脉冲星候选体诊断图分区压缩,使用近年来FAST巡天搜索项目脉冲星候选体诊断图来训练和验证。结果表明,改进的跳白块编码(White Block Skipping,WBS)压缩稀疏黑白图像的性能是PNG(Portable Network Graphics)的5倍;深度网络压缩算法处理灰度图和彩色图峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)性能优于JPEG(Joint Photographic Experts Group)和JPEG2000算法,与BPG(Better Portable Graphics)算法性能相当,结构相似性(Structural Similarity,SSIM)远超传统压缩算法。The magnitude of candidate diagnostic images generated by FAST pulsar search has increased exponentially, which brings challenges to scientific data management. It is urgent to study compression methods to achieve effective storage of diagnostic images and speed up their transmission and sharing in the network. It is not reasonable to simply treat the whole image as a color image and compress it with the same method. We propose a skip white block coding and a deep network compression coding model to compress pulsar candidate diagnostic images. We take the pulsar candidate diagnostic images from recent FAST sky survey search projects to train and verify the model. The results show that the performance of the improved WBS compression for sparse black and white images is 5 times than PNG. The performance of deep network compression algorithm for PSNR is better than JPEG and JPEG2000, comparable to BPG. And it is much better than traditional compression algorithm for SSIM.

关 键 词：候选体诊断图压缩 深度网络压缩模型 跳白块编码 500 m口径球面射电望远镜 

分 类 号：P162[天文地球—天文学]