Application of Real-ESRGAN in improving IR sensor Images for use in SAR operations

Abstract

The use of Unmanned Aerial Vehicles (UAVs) in search and rescue operations has grown significantly, primarily due to reduced costs and lower associated risks. However, the effectiveness of these vehicles is closely linked to the quality of the sensors used for target capture and identification, making the investigation of these devices a crucial area of research. This study presents a systematic review of the literature on the application of Generative Adversarial Networks (GANs) in UAV-generated images, with a focus on search and rescue. Additionally, we introduce a methodology that uses the Real-ESRGAN tool to enhance images obtained by UAVs during search and rescue missions, specifically targeting sensors that operate in the infrared spectrum. The results of applying this technique to our dataset show significant improvements in image quality, suggesting that this approach may have valuable applications in post-processing and in the identification of human targets in search and rescue operations.