Design of low-light image object detection algorithm based on improved YOLO model

Due to insufficient lighting, images captured in low-light conditions often exhibit significant noise, low contrast, and blurring. These degradations lead to considerable performance degradation when conventional object detection methods are applied in such environments. To address these challenges, this paper introduces an enhanced object detection algorithm based on the YOLOv11n architecture, specifically designed to improve detection performance in low-light scenarios. The proposed approach incorporates two key improvements. First, an iAFF attention mechanism is integrated into the backbone network. This module adaptively recalibrates the fusion process across different feature maps, allowing the model to dynamically adjust feature importance according to both image content and task requirements. Second, Dysample upsampling is employed to replace standard upsampling operations. This technique enhances flexibility and efficiency by dynamically selecting upsampling strategies tailored to different regions of the feature map. Experimental results show that after introducing both iAFF and Dysample, the mAP50 and mAP50:95 of the YOLOv11n algorithm increase by 4.03% and 2.29% respectively compared to the original YOLOv11n before improvement, indicating a significant enhancement in the recognition ability of low - light - level images.