LDA-YOLO: a lightweight deblurring-aware network for real-time object detection in blurred aerial images

Accurate and real-time object detection in aerial images remains a challenging task due to diverse capture conditions, including motion blur, defocus, scale variation, and complex backgrounds. These issues are further exacerbated when dealing with small or occluded objects commonly found in low-altitude UAV footage, satellite imagery, and vehicle-mounted aerial systems. In this article, we propose a lightweight deblurring-aware object detection network, termed LDA-YOLO, designed to achieve robust and efficient detection across a wide range of aerial imaging platforms. The proposed architecture integrates a fast deblurring module to restore structure from degraded inputs, a dual-domain feature aggregation module (DDFA) to enhance feature focus on informative regions, and a multi-scale feature fusion network based on lightweight backbones to support small object detection under scale and blur variations. A YOLOv8-style detection head enables low-latency prediction suitable for edge deployment. Comprehensive experiments on four representative aerial datasets—VisDrone, DOTA, Drone Vehicle, and VEDAI—demonstrate that LDA-YOLO achieves competitive accuracy while maintaining real-time performance on embedded platforms, offering a deployable and effective solution for general-purpose aerial image object detection under degraded visual conditions.