A novel approach to atmospheric turbulence detection in free-space optical communication links using DCGAN with hyperparameter optimization and transfer learning models

This study focuses on detecting turbulence-induced disturbances in Free-Space Optical (FSO) communication links, which are highly susceptible to environmental factors such as temperature, pressure, and humidity variations. To address the problem of limited data in turbulence detection, we applied DCGAN-based data augmentation to a dataset in this area for the first time in the literature, increasing the sample size and achieving class balance across six different turbulence types. Bayesian Optimization was used to fine-tune DCGAN hyperparameters, leading to high-quality synthetic images as validated by Inception Score (IS) and Fre<acute accent>chet Inception Distance (FID) metrics. We trained ResNet-50, EfficientNetB7, DenseNet121 and InceptionV3 transfer learning models on the augmented dataset, and ResNet-50 achieved the best balance between computational efficiency with 98.89 % accuracy and 7789 ms training time. Experimental results highlight the potential of deep learning in improving FSO communication reliability under diverse atmospheric conditions. Future work could focus on collecting larger, more varied real-world datasets, employing advanced augmentation techniques, optimizing models for real-time processing, and integrating FSO systems with emerging technologies like 5G/6G and LIDAR to enhance robustness against environmental disturbances.