LAE-GSDetect: A Lightweight Fusion Framework for Robust Small-Face Detection in Low-Light Conditions

In response to the challenges of insufficient accuracy in face detection and missed small targets under low-light conditions, this paper proposes a detection scheme that combines image preprocessing and detection model optimization. Firstly, Zero-DCE low-light enhancement is introduced to adaptively restore image details and contrast, providing high-quality inputs for subsequent detection. Secondly, YOLOv11n is enhanced through the following improvements: a P2 small-target detection layer is added while the P5 layer is removed, addressing the original model's deficiency in detecting small targets and streamlining the computational process to balance model complexity and efficiency; the P2 upsampling is replaced with DySample dynamic upsampling, which adaptively adjusts the sampling strategy based on features to improve the accuracy of feature fusion; a lightweight adaptive extraction module (LAE) is incorporated to reduce the number of parameters and computational costs; finally, the detection head is replaced with GSDetect to maintain accuracy while reducing computational overhead. Experimental results show that the improved model reaches an mAP50 of 58.7%, which is a 10.2% increase compared with the original model. Although the computational complexity increases to 7.5 GFLOPs, the parameter count is reduced by 45%, offering a more optimal solution for face detection in low-light environments.