基于改进RetinaNet模型的室内安全帽佩戴检测研究

Research on Indoor Hard Hat Wear Detection Based on Revised RetinaNet Model

在生产车间内,生产情况较为复杂,在生产过程中工人未按要求佩戴安全帽是造成安全事故的原因之一。目前基于室内场景的安全帽佩戴检测研究还较少,现提出一种改进的RetinaNet网络检测模型,用于生产车间内员工安全帽佩戴检测。首先,该模型通过在ResNet-50上采用卷积块注意力模块,在其中的通道注意模块的MLP网络中添加dropout机制,增强模型的泛化性;然后,采用K-MeansⅡ维度聚类算法找出锚点的合适尺度来进行目标的检测。实验结果显示,改进的模型在不同实验条件下,安全帽佩戴检测精度达到98.71%,检测速率达到15.6f/s,能满足生产车间的实际需求。

In the production workshop,the production situation is complicated,and the failure of workers to wear safety helmets as required in the production process is one of the causes of safety accidents.At present,there are few researches on helmet wearing detection based on indoor scenes.An improved RetinaNet network detection model is proposed for helmet wearing detection of employees in production workshops.Firstly,the model adopts convolution block attention module in ResNet-50,and adds dropout mechanism in MLP network of channel attention module to enhance the generalization of the model;Then,K-means dimension clustering algorithm is used to find the appropriate scale of anchor point for target detection.The experimental results show that,the accuracy of helmet wearing detection reaches 98.71%and the detection rate reaches 15.6f/s under different experimental conditions,which can meet the actual needs of the production workshop.