Object detection and recognition in foggy environment based on improved YOLOv11

Aiming at the performance degradation of existing target detection models in foggy environments and the insufficient processing capability of traditional YOLOv11 network for degraded features of foggy images, this paper proposes an improved model based on the YOLOv11 network ---- ICG_YOLOv11, which aims to improve the target detection accuracy in foggy environments. The image improvement of fog images is carried out through the Illumination Adaptive Transformer framework, and at the same time, joint training with the detection network is performed to learn the correlation features between fog and detected targets; secondly, the Cascaded Group Attention attention mechanism is incorporated into the YOLOv11 neck to improve the feature extraction capability. The experimental results show that the proposed model has excellent performance on the Foggy Cityscapes dataset, which effectively improves the detection accuracy of the model in foggy environments. Compared with the benchmark model, the average accuracy (mAP@50) at IOU threshold 0.5 is improved by $0.8 \%$ and $2.9 \%$, respectively.