LLIE-YOLO: A Task-Driven Joint Enhancement and Detection Optimization Method for Low-Light Object Detection

To address the problems of insufficient illumination, noise interference, and missing texture details in low-light object detection, this paper proposes a joint enhancement-detection optimization framework, named LLIE-YOLO. Built on YOLOv8, the proposed method introduces adaptive Gamma correction for preprocessing, a task-driven enhancement module AttnSCI-CEN, a texture-aware BiFPN, and a quality-aware joint loss AQA-Loss to improve feature representation and detection performance in low-light scenes. Experimental results on the ExDark dataset show that the proposed method achieves an [email protected] of 0.809, which is 5.7 percentage points higher than that of the baseline YOLOv8l, with only limited computational overhead. The results demonstrate that LLIE-YOLO can effectively improve low-light object detection while maintaining efficiency.