Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Yıldız Teknik Üniversitesi, Elektrik-Elektronik Fakültesi, Elektronik Ve Hab.Müh.Böl, Türkiye
Tezin Onay Tarihi: 2025
Tezin Dili: İngilizce
Öğrenci: Erman EKŞİ
Danışman: Murat Taşkıran
Özet:
Thermal imaging plays a pivotal role in diverse applications, such as surveillance, medical diagnostics, and industrial monitoring. However, its utility is often constrained by the low resolution of thermal images, which limits their effectiveness in tasks requiring detailed visual analysis. This research introduces an enhanced super-resolution (SR) framework tailored to thermal imaging, leveraging improvements to the SwinIR (Swin Transformer for Image Restoration) algorithm. In the SwinIR architecture, the Gaussian Error Linear Unit (GeLU) activation function is substituted with the Rectified Linear Unit (ReLU) to improve visual output quality and optimize computational performance. To establish a robust foundation, the study reviews and evaluates various existing SR methodologies, including interpolation-based, reconstruction-based, and learning-based approaches. The SwinIR-ReLU model, modified for this study, is evaluated on a high-resolution thermal image dataset. Its performance is assessed using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM). Experimental results reveal that the proposed model surpasses the original SwinIR-GeLU configuration, delivering superior PSNR and SSIM values. Additionally, qualitative analysis highlights improved edge detail preservation and structural clarity, reinforcing its suitability for practical deployment. This work advances the domain of image super-resolution by presenting an optimized solution for thermal imaging challenges. Future research could investigate further enhancements to the SwinIR architecture, explore alternative activation functions, or integrate generative adversarial networks (GANs) to amplify the model's capabilities. The results highlight the significant impact of transformer-based architectures in improving the resolution of low-quality thermal images.