A laser beam carrying object information will become a random speckle pattern as it propagates through a hollow core fiber (HCF) of large inner diameter. We propose and experimentally demonstrate a technique that employs the convolutional neural network (CNN), which is one type of deep learning, for image reconstruction from the random speckle pattern. In the experimental demonstration, 4000 speckle images and 4000 corresponding original images are used to train CNN, and 1000 speckle images are used to test the trained CNN. The experimental results demonstrate the possible realization of deep learning for the image reconstruction through the HCF. Furthermore, the influence of the diameter and length of HCF on the fidelity of the image reconstruction is quantitatively investigated by determining the accuracy and Pearson correlation coefficient (PCC).

携带目标信息的激光束通过大直径的空心光纤（HCF）传播时，将成为随机的斑点图案。我们提出并通过实验证明了采用卷积神经网络（CNN）的技术，该技术是一种深度学习，用于根据随机斑点图案重建图像。在实验演示中，使用4000个斑点图像和4000个相应的原始图像来训练CNN，并使用1000个斑点图像来测试训练后的CNN。实验结果证明了通过HCF实现深度学习用于图像重建的可能。此外，