摘要
新冠病毒在全球传播期间,规范佩戴口罩是最有效的防范方式。对公共场所中密集人群的口罩佩戴是否规范进行检测时,由于目标紧邻、遮挡以及含有大量的小目标,存在检测精度低、错检和漏检率高的问题。为了解决上述问题,文章提出一种基于改进RetinaNet模型的口罩规范佩戴检测方法。通过引入ECA-Net注意力模块,使得对口罩目标特征给予更多的关注,提高检测精度;其次,在特征金字塔FPN后引入自适应空间特征融合模块ASFF,来充分利用多尺度特征,进行更加充分的融合。使用该文所提出的方法在自制的口罩规范佩戴数据集进行实验,结果表明该文方法的整体性能优于其他的检测算法。
During the global spread of COVID-19,regulating the wearing of masks is the most effective form of prevention.When detecting whether masks are worn properly by dense groups of pcople in public places,there are problems of low detection accuracy and high rates of misdetection and omission due to the close proximity of targets,occlusion and the presence of a large number of small targcts.In order to solvc the abovc problcms,this papcr proposcs a mask spccification wearing detection method based on improved RetinaNet model.The introduction of the ECA-Net Attention Module makes it possible to give more attention to thc mask target features and improve the detection accuracy;Secondly,the adaptive spatial feature fusion module ASFF is introduced after the feature pyramid FPN to make full usc of the multi-scale featurecs so that they can be more fully fused.Experiments using the method proposed in this paper on a homemade mask specification wearing dataset show that the overall performance of the method in this paper outperforms other detection algorithms.
作者
张思甜
刘军清
康维
ZHANG Sitian;LIU Junqing;KANG Wei(College of Computer and Infomation Technology,China Three Gorges University,Hubei Yichang 443002,China)
出处
《长江信息通信》
2024年第2期35-38,共4页
Changjiang Information & Communications
