SRGAN Enhancement through Autoencoder-Pretrained U-Net with Residual Blocks for Improved Image Super-Resolution | ||
| Journal of Modeling and Simulation in Electrical and Electronics Engineering | ||
| دوره 4، شماره 3 - شماره پیاپی 17، زمستان 2024، صفحه 11-24 اصل مقاله (1.5 M) | ||
| نوع مقاله: Research Article | ||
| شناسه دیجیتال (DOI): 10.22075/mseee.2025.36700.1198 | ||
| نویسندگان | ||
| Amirreza Rouhbakhshmeghrazi* 1؛ Bo Li2؛ Shayan Nalbandian3؛ Chao Song4؛ Ghazal Alizadeh5؛ Mohammad Reza Hassannezhad6 | ||
| 1School of Electronics and Information, Northwestern Polytechnical University, Xi'an, China. | ||
| 2School of Electronics & Information, Northwestern Polytechnical University, Xi’an, Shaanxi, China. | ||
| 3School of Software Engineering, Northwest Polytechnic University, Xi'an, China. | ||
| 4School of Electronics & Information, Northwest Polytechnic University, Xi'an, China. | ||
| 5School of Aeronautics, Northwestern Polytechnical University, Xi'an, China. | ||
| 6School of Aeronautics, Northwestern Polytechnical University, Xi’an, Shaanxi, China. | ||
| چکیده | ||
| Super-resolution is a crucial task in image processing, enhancing the resolution of low-quality images for applications such as surveillance, remote sensing, and autonomous systems. Traditional methods often struggle to preserve fine details, leading to artifacts and reduced visual fidelity. This study introduces the Pretrained RU-SRGAN, an enhanced Super-Resolution Generative Adversarial Network (SRGAN) that incorporates U-Net architecture, residual learning, and autoencoder pretraining to improve both image quality and computational efficiency, particularly in resource-constrained environments like UAVs. The goal of this research is to evaluate how these architectural modifications can enhance super-resolution performance with limited data. Autoencoder pretraining enables the generator to leverage learned features from low-resolution images, accelerating convergence and improving high-resolution reconstructions. Experimental results show that Pretrained RU-SRGAN outperforms baseline models, achieving a PSNR of 25.7 dB and an SSIM of 0.83. These results highlight the model's ability to preserve fine details and structural integrity, making it particularly effective for real-time image enhancement in UAV applications. The Pretrained RU-SRGAN provides a robust solution for super-resolution tasks, balancing high-quality image reconstruction with computational efficiency, and is well-suited for practical deployment in dynamic, resource-limited environments. | ||
| کلیدواژهها | ||
| Transfer Learning؛ UAV؛ Image Reconstruction؛ RU-Net؛ SSIM؛ PSNR | ||
| مراجع | ||
|
| ||
|
آمار تعداد مشاهده مقاله: 557 تعداد دریافت فایل اصل مقاله: 558 |
||
| تعداد نشریات | 22 |
| تعداد شمارهها | 723 |
| تعداد مقالات | 10,402 |
| تعداد مشاهده مقاله | 72,899,734 |
| تعداد دریافت فایل اصل مقاله | 64,575,435 |