【摘要】 针对目标检测算法在雾天天气下的识别精度低、容易漏检等问题，提出一种基于YOLOv5框架的先去雾后检测的算法。利用改进的AOD-Net去雾算法进行图像增强，采用Twin-transform Block模块替换提取中小目标特征，同时添加CBMA注意力机制，增强模型提取图像特征的能力。采用DIoU＿NMS后处理方法，精准回归预测框。实验结果表明，在BDD100K数据集上进行对比分析，相较于原有YOLOv5s网络，文中提出的改进方案准确率明显提升，其中薄雾场景中mAP提高1.59%,浓雾场景中mAP提高5.80%。因此，文中提出的算法能够有效地改善雾天条件下目标识别中存在的漏检问题。更多还原

【Abstract】 Aiming at the problems of low recognition accuracy and easy missed detection of target detection algorithms in foggy weather, a new algorithm based on the YOLO v5 framework is proposed to remove fog first and detect later. In this paper, the improved AOD-Net defogging algorithm is used for image enhancement; Use the Twin-transform Block module to replace and extract small and medium target features; At the same time, the CBMA attention mechanism was added to enhance the model’s ability to extract image features. Adopting DIoU＿ NMS post-processing method, accurate regression prediction box. The experimental results show that compared to the original YOLOv5s network, the accuracy of the proposed improvement scheme is significantly improved on the BDD100K dataset, with an increase of 1. 59% in the haze scenario and 5. 80% in the dense fog scenario. Experimental results show that the algorithm proposed in this paper can effectively improve the problem of missed detection in target recognition under foggy conditions.更多还原