针对上海光源为代表的科学大装置产生的超大规模科学数据流的传输和处理瓶颈，结合前期参与国际SKA望远镜和中国500米口径球面射电望远镜(FAST)中积累的预处理与协同设计研究方法，提出研究面向像素级极小尺寸目标的神经元网络筛选和基于可变精度计算压缩方法，形成面向光源光栅成像投影和衍射数据的预处理压缩方法；研究基于下一代网络的网络接入方法，形成超大规模科学数据流的流量卸载和网内缓存；研究图像重建算法的并行化，形成基于图形处理器（GPU）和现场可编程门阵列（FPGA）的超大规模图像重建算法并行加速；进一步在光源线站现场面向超大规模数据流进行验证和优化，尝试突破超大规模科学数据流的传输和处理瓶颈，加快我国取得重大科学发现的进展，推进张江国家实验室的建设。

To handle the bottlenecks of transmission and processing of super-large scientific.data streams generated by huge scientific apparatuses such as Shanghai Synchrotron Radiation Facility (SSRF), with the preprocessing and co-design methods gained in square kilometer array radio telescope (SKA) project and five-hundred aperture sphere telescope (FAST) project, we propose to optimize the neural network filtering of tiny objects at pixel level and variable precision computing and compression in the form of preprocessing and compression methods for raster projection and diffraction images from SSRF. We also propose to design a write-in method for next generation network in order to achieve traffic offloading and network storage. We further propose to work on parallelization of image reconstruction by accelerating on graphic processing unit (GPU) and field programmable gate array (FPGA). Our design and implementation will be validated on-site at stations of SSRF for further optimization to resolve the bottlenecks of transmission and processing of super-large scientific data streams and accelerate the progress of major scientific discoveries in China. These efforts will be also helpful to accelerate the progress in Zhangjiang National Laboratory.