Foggy Weather Scene Object Detection Algorithm: FREFog-Yolov8s

High-accuracy object detection in foggy conditions is crucial for autonomous driving and intelligent surveillance. However, fog reduces contrast and blurs details, degrading the performance of traditional detection models. While some models are optimized for foggy conditions, they often require high computational resources, limiting deployment efficiency. To address this, we propose FREFog-YOLOv8, which enhances detection accuracy while reducing computational cost. First, we introduce the FusedSPPFAVG module, which integrates average and max pooling to improve global feature perception. Second, we propose CSP-REP, combining multi-branch feature fusion with RepbitBlock to enhance multi-scale feature extraction. Third, we integrate the EMA attention mechanism into RepVitBlock, achieving dual-stage optimization for better feature representation. Finally, ShapeIOU loss is used to refine bounding box matching accuracy. On the RTTS dataset, our method achieves mAP@ 0.5 of 80.5 and mAP@ 0.5: 0.95 of 58.5, surpassing both the baseline and state-of-the-art models while reducing parameter size and computation cost, balancing accuracy and efficiency.