SPRNet: laser spot center position and reconstruction under atmospheric turbulence based on deep learning enhancement

Optical communication suffers from atmospheric turbulence for free space optical communication (FSOC) and the received spot has undergone severe wavefront distortion. It is difficult to position the spot center accurately or reconstruct the original spot, which leads to the loss of the transmitted information. Therefore, we establish a novel neural network to achieve spot center position and reconstruction, named SPRNet. Our SPRNet consists of spot structural feature extraction (SSFE) module and field distribution feature enhancement (FDFE) module to locate the center and restore the quality-enhanced spot. In FDFE module, we propose a novel spot-constrained attention module to better fuse the dual feature. To solve the problem of lacking ground truth (label), we propose the multi-frame aggregation method to obtain the labels to train our deep-learning-based method and establish the Turbulence50 dataset. We carried out experiments with simulated data and real-world data to verify the effectiveness of our SPRNet. The experiment results show that our method has better performance and strong robustness compared to other methods, which improves more than 2.2422 pixels on the benchmark of Manhattan distance for spot center position and more than 3.2477dB on the benchmark of PSNR for spot reconstruction.