基于注意力机制的YOLOv5算法在跑道异物检测中的应用

Application of YOLOv5 algorithm based on attention mechanism in foreign object debris detection

机场跑道异物严重影响飞行安全,针对现有算法对小目标存在误检、漏检等问题,提出一种改进的YOLOv5算法对机场跑道异物进行检测。在YOLOv5的主干网络中添加有效通道注意力(ECA)模块,通过少量参数的增加带来明显的性能增益。将颈部网络中原特征金字塔模块替换为加权双向特征金字塔(BiFPN)网络,实现双向跨尺度连接和加权特征融合。采用EIoU Loss作为损失函数,加快了收敛速度。在FOD⁃A数据集上的实验表明,改进后的YOLOv5模型均值平均精度(mAP@0.5)指标达到了97.4%,相比于原模型提高了1.6个百分点。

The foreign objects debris on airport runways seriously affect the flight safety,and an improved YOLOv5 algorithm is proposed to detect foreign objects on airport runways to address the problems of false detection and missed detection of small tar⁃gets in existing algorithms.The efficient channel attention(ECA)module is added to the backbone network of YOLOv5,which brings obvious performance gain by a small increase of parameters.The original feature pyramid module is replaced with a weighted bi-directional feature pyramid network(BiFPN)in the neck network to achieve bi-directional cross-scale connectivity and weighted feature fusion.The EIoU Loss is used as the loss function to accelerate the convergence speed.Experiments on the FOD-A dataset show that the mean average accuracy mAP@0.5 metric of the improved YOLOv5 model reaches 97.4%,which is 1.6%better than that of the original model.