Local Adaptive Illumination-Driven Input-Level Fusion for Infrared and Visible Object Detection

Remote sensing object detection based on the combination of infrared and visible images can effectively adapt to around-the-clock changeable illumination conditions. However, most existing networks need two backbone extract features modalities, respectively. Compared with single modality network, this greatly increases amount calculation, which limits its real-time processing vehicle unmanned aerial (UAV) platforms. Therefore, paper proposes a local adaptive illumination-driven input-level fusion module (LAIIFusion). The previous methods for perception only focus global illumination, ignoring differences. In regard, we design new submodule, newly define value illumination. With more accurate area selection label design, perceive scene condition. addition, aiming at problem incomplete alignment between images, submodule is designed rapid estimation slight shifts. experimental results show that algorithm LAIIFusion ensure large improvement in accuracy small loss speed. On DroneVehicle dataset, our combined YOLOv5L could achieve best performance.