Research on foreign object detection on highways based on improved YOLO in harsh weather

Aiming at the problem of difficult recognition of foreign object intrusion on highways under harsh weather, an improved YOLOv8 detection algorithm is proposed to enhance the accuracy of road foreign object target detection. More refined feature extraction is achieved through the SE attention mechanism, which improves the detection accuracy in the case of image blurring as well as fog and rain line occlusion in harsh weather; the backbone network improved by the sensory wild attention convolution RFA is used to enhance the ability to extract features from low-resolution and small targets; and the use of regression loss combining with focusing loss using SIoU further strengthens the network's classification and localisation ability. Experimental results show that the improved model performs well on the harsh weather traffic obstacle detection dataset, with a mAP50 of 86.8% outperforming the original YOLOv8 network at 84.2%.