Hybrid deep learning framework for image restoration in Fso systems affected by log-normal fading

Image degradation caused by noise and atmospheric fading poses a significant challenge to the reliability and performance of image transmission systems, particularly in Free Space Optical (FSO) communication employing high-order modulation schemes like Quadrature Amplitude Modulation (64-QAM). In such systems, log-normal fading is a result of atmospheric turbulence can severely degrade image quality, reducing the Peak Signal-to-Noise Ratio (PSNR) and impairing critical visual information. Traditional mitigation techniques such as Multiple-Input Multiple-Output (MIMO), spatial diversity, adaptive optics, and filtering methods (e.g., Gaussian and Wiener filters) offer partial improvements but often fall short in dynamic or nonlinear environments. To address these limitations, this paper proposes a hybrid deep learning-based image restoration framework that integrates a Deep Convolutional Neural Network (DCNN) with post-processing image sharpening techniques. The DCNN is first trained on a synthetically generated dataset of 30,000 images, incorporating both faded and non-faded samples across a wide range of Signal-to-Noise Ratio (SNR) values (0-40 dB), to effectively learn and suppress the statistical effects of log-normal fading. Following the denoising stage, an image sharpening method is applied to enhance clarity and perceptual quality. Experimental evaluations demonstrate a substantial improvement in image quality, with PSNR rising from as low as 19 dB to 68 dB, outperforming conventional methods in both quantitative metrics such as PSNR and the Root Mean Square Error (RMSE), and qualitative visual assessments. The proposed approach not only restores structural image content with high fidelity but also proves highly effective for real-world applications requiring accurate image interpretation, such as medical imaging, remote sensing, and surveillance under atmospheric distortion.