基于循环一致性对抗网络的稀疏阵列宽带合成波束效应消除

Wideband Beam Effect Elimination of Sparse Array Based on Cyclic Consistency Adversarial Network

射电天文稀疏干涉阵列成像过程中由于无线电接收器的带宽限制会使观测后的图像产生宽带合成波束效应,针对这个问题,本文设计了一个带宽合成波束效应的消除CycleGAN模型。该模型利用CycleGAN模型中的残差学习机制对具有空间复杂结构的射电源信号携带的带宽合成波束效应图像进行特征提取,进而提高恢复效果。通过天文通用软件CASA模拟出来的脏图和原图作为图像对模型进行训练。这种结合方式能够有效地将两种图像风格进行转换,从而使得模型能够更好地适应不同的射电源信号。实验结果显示,该深度学习算法与现有的宽带合成波束算法在图像指标PSNR和SSIM上得到明显提升,能够有效地恢复天空图像,这一技术将为我们提供更为准确的天文学数据,并推动天文学的发展。

Due to the bandwidth limitation of the radio receiver, wideband beam effect is generated in the imaging of radio astronomy sparse interference array. To solve this problem, a CycleGAN model is designed to eliminate the beam effect. In this model, the residual learning mechanism of CycleGAN model is used to extract the features of bandwidth-synthesised beame ffect images car-ried by radio source signals with complex spatial structure, so as to improve the recovery effect. The model was trained by using the dirty map and the original image simulated by CASA—a astronomical software. This combination method can effectively convert the two image styles, so that the model can better adapt to different radio source signals. The experimental results show that the deep learning algorithm is significantly improved in PSNR and SSIM compared with the existing wideband synthetic beam algorithm, and can effectively restore the sky image. This technology will provide us with more accurate astronomical data and promote the development of astronomy.