基于Yolov5改进的香烟目标检测

Improved cigarette target detection based on Yolov5

因为吸烟引起的火灾,给人民群众生活带来了巨大伤害,针对禁烟场所香烟目标检测精度不高的问题,文章提出了一种基于Yolov5改进的目标检测模型。该模型首先在Yolov5的颈部网络中引入SimAM(Simple,Parameter-Free Attention Module)无参注意力机制,加强Yolov5的特征融合能力。其次是使用SIOU(Soft Intersection OverUnion)作为定位回归框的损失函数,提高边界框的定位精度。最终将Yolov5和改进后的模型在自建数据集作对比实验,结果表明,该模型的准确率达到89.6%,召回率(Recall)达到81.4%,平均精度(MAP)达到84.3%,与原本的Yolov5模型相比,检测效果均有了提升。

Due to the fire caused by smoking,it has caused great harm to the lives of the people.In response to the problem of low accuracy in cigarette target detection in non-smoking areas,this paper proposes an improved object detection model based on Yolov5.This model first introduces the SimAM(Parameter Free Attention Module)mechanism in Yolov5's neck network to enhance its feature fusion ability.Secondly,SIOU(Soft Intersection Over Union)is used as the loss function of the positioning regression box to improve the positioning accuracy of the bounding box.Finally,YOLOv5 and the improved model were compared in a self built dataset.The results showed that the accuracy of the model reached 89.6%,the recall rate reached 81.4%,and the average accuracy(MAP)reached 84.3%.Compared with the original YOLOv5 model,the detection performance was improved.