改进YOLOv5s的X光安检图像违禁物品检测算法

Detection of threat items in X-ray security inspection images based on improved YOLOv5s algorithm

针对现有X光安检图像中违禁物品检测精度低的问题,基于YOLOv5s(you only look once version 5 small)提出了一种改进的违禁物品检测算法。利用重参数思想设计了一种Rep模块以协助YOLOv5s主干网络提取更多特征信息,在不增加推理时间的基础上提高算法检测精度。同时,在YOLOv5s颈部的路径聚合网络中插入2个通道注意力机制压缩-激励模块,加强通道间的相关性,提高整体网络的检测效果。在SIXray数据集上的实验结果表明,在不增加检测时间的基础上,改进的YOLOv5s算法比原始算法在平均精度均值(mAP)、宏精确率(macro precision)、宏召回率(macro recall)和宏F1(macro-F1)这4个评价指标上分别提升了2.6、2.0、4.0和3.0个百分点。

In view of the problems of slow recognition efficiency and low detection accuracy of threat items in the existing X-ray security inspection images,an improved threat item detection algorithm is proposed based on you only look once version 5 small(YOLOv5s).A rep module is designed based on the idea of reparameterization to assist the YOLOv5s backbone network in extracting more feature information,improving the detection accuracy of the algorithm without increasing the inference time.Moreover,two channel attention mechanism SE blocks are inserted into the path aggregation network of the neck of YOLOv5s to strengthen the correlation between channels and improve the detection effect of the whole network.The experimental results on the SIXray dataset show that,compared to the original algorithm,the improved one improves the metrics of mean average precision,macro precision,macro recall,and macro F1 by 2.6,2.0,4.0,and 3.0 percent,respectively.