Atmospheric turbulence recognition with deep learning models for sinusoidal hyperbolic hollow Gaussian beams-based free-space optical communication links

Abstract The integration of artificial intelligence technology to improve the performance of free-space optical communication (FSO) systems has received increasing interest. This study aims to propose a novel approach based on deep learning techniques for detecting turbulence-induced distortion levels in FSO communication links. The deep learning-based models improved and fine-tuned in this work are trained using a dataset containing the intensity profiles of Sinusoidal hyperbolic hollow Gaussian beams (ShHGBs). The intensity profiles included in the dataset are the ones of ShHGBs propagating for 6 km under the influence of six different atmospheric turbulence strengths. This study presents deep learning-based Resnet-50, EfficientNet, MobileNetV2, DenseNet121 and Improved+MobileNetV2 approaches for turbulence-induced disturbance detection and experimental evaluation results. In order to compare the experimental results, an evaluation is made by considering the accuracy, precision, recall, and f1-score criteria. As a result of the experimental evaluation, the average values for accuracy, precision, recall and F-score with the best performance of the improved method are given; average accuracy 0.8919, average precision 0.8933, average recall 0.8955 and average F-score 0.8944. The obtained results have immense potential to address the challenges associated with the turbulence effects on the performance of FSO systems.