Real-Time Smoke Removal and Rescue System Using HSRDN

Abstract

Fire accidents have become increasingly severe in recent years due to heavy smoke, which significantly reduces visibility and delays rescue operations. Poor visual conditions limit the effectiveness of vision-based rescue systems, making real-time smoke removal and accurate victim detection essential. Existing smoke removal and image enhancement methods, including traditional dehazing and deep learning models such as FFA-Net, mainly focus on visual enhancement and suffer from high computational complexity, reduced performance in dense smoke, and lack of intelligent rescue decision-making, while object detection models like YOLO experience accuracy degradation under low-visibility conditions. To overcome these limitations, this paper proposes an AI-driven Real-Time Smoke Removal and Rescue System using HSRDN, which integrates real-time smoke removal, enhanced object detection, and fuzzy logic–based rescue prioritization. The performance of the proposed system is evaluated using PSNR and SSIM to measure image quality and structural restoration, mAP to assess detection accuracy, and FPS to validate real-time processing efficiency. Experimental results demonstrate that the proposed approach provides improved visibility, higher detection accuracy, and reliable real-time performance, making it suitable for practical fire rescue applications.