Research On High Dynamic Range Reconstruction Of SKA Images Based On Big Data Method

With the construction of the international mega-science project Square Kilometre Array(SKA),high-fidelity and high-dynamic-range imaging has become increasingly important.However,due to incomplete sampling in the uv plane,sidelobes appear in the point spread function,which interfere with the imaging process and cause weak signal sources to be submerged beneath the sidelobes,thereby limiting the dynamic range of the images.To address this issue and improve the accuracy and dynamic range of imaging,this study develops and trains a conditional generative adversarial network(GAN)that combines perceptual loss and attention mechanisms to learn the mapping between dirty images and real radio sky images in the SKA imaging process.The model improves the generator by introducing a deep residual network and attention mechanism,uses the Patch GAN architecture as the discriminator,and selects appropriate normalization operations,activation functions,and optimizers for practical image tasks to enhance the accuracy of the model.Additionally,to preserve the content information of the original images,the model also incorporates perceptual loss.To validate whether the model can reconstruct the real sky brightness distribution from dirty images and recover faint sources from the vicinity of bright sources,reconstruction experiments on different source structures and comparison experiments with the commonly used CLEAN algorithm are conducted.The results demonstrate that the model exhibits strong reconstruction capability in low-frequency SKA simulated data,enabling high-fidelity and high-dynamic range reconstruction of SKA images.Compared to traditional deconvolution methods,the model can improve the dynamic range by 1 to 4 times and achieve the reconstruction of faint sources from very bright sources within a larger data range.Meanwhile,the model significantly improves the PSNR,RMSE,and SSIM of the reconstructed images,resulting in more accurate and detailed sky images.This study aims to address the problem of high-fidelity and high-dynamic range image reconstruction in the next-generation radio telescope SKA,and it is the first research on SKA image reconstruction based on deep learning and big data patterns.The research has certain reference value for SKA data processing and high-dynamic range imaging.This study focuses on addressing the problem of high-fidelity and high-dynamicrange imaging in the next-generation radio telescope SKA,and presents the first research on SKA image reconstruction based on deep learning and big data modeling.The findings of this study have certain reference value for SKA data processing and high-dynamic-range imaging.