一种Transformer引导YOLOv5对高空吊钩违规操作识别

A Transformer Guides YOLOv5 to Identify Illegal Operation of High-altitude Hooks

为了降低高空作业意外事故发生的风险,高空吊钩作业违规检测识别并及时告警显得十分重要。针对高空吊钩目标小以及图像目标模糊等问题,现有的基于YOLOv5的目标检测方法存在误检测率高、效率低等问题,为此采用一种Transformer模块来引导YOLOv5对高空吊钩违规操作检测的算法。首先,在Backbone中添加Transformer模块有效捕获全局信息和目标图像的上下文内容信息,有利于捕获复杂背景干扰下目标的特征信息;其次在Neck中使用BiFPN模块,能有效挖掘小目标图像内容信息及深层图像语义信息;最后使用SIoU损失函数,可以更准确定位目标框的位置信息。

In order to reduce risk and accidents in high-altitude operations,it is very important to detect and identify illegal operations and warn timely.Due to small size and blurry of the hooks′image,the existing object detection method based on YOLOv5 has high false detection rate and low efficiency.In this paper,a Transformer module is proposed to guide YOLOv5 to detect the violative behaviors in high-altitude hooks′usage.Firstly,the Transformer module is added to Backbone to effectively capture the global information and the contextual content information of the target image,which is conducive to capturing the feature information of the target under complex background interference.Secondly,the BiFPN module is used in Neck,which can effectively mine the content information of small target images and deep image semantic information.Finally,the SIoU loss function can be used to locate the position information of the target frame more accurately.