North China Institute of Science and Technology

针对基于神经网络的安全帽检测工作场景模型部署嵌入式设备困难的问题,提出了一种基于改进YOLOv5s的轻量化目标检测算法。首先将YOLOv5s的主干网络替换为轻量化网络MobilenetV3,降低模型的参数量和计算量,保留模型的SPPF模块以提高模型对不同尺寸目标的检测能力;其次,在模型的Neck和Head之间添加注意力机制,以更好地捕获图像中的目标信息,提高精确度和鲁棒性;最后,将模型损失函数替换为EIoU,加速模型收敛,提高模型检测精度。通过自建安全帽数据集对所提模型进行了训练和验证,实验结果表明,相较于原模型,检测精确度提升了1.2%,参数量降低了39.4%、计算量降低了56.3%,模型体积压缩了38.6%,为基于改进YOLOv5s的安全帽识别算法在嵌入式设备上部署提供了一种有效的方法。

A lightweight target detection algorithm based on improved YOLOv5s is proposed to address the difficulty of deploying embedded devices for the neural network-based helmet detection work scenario model. Firstly, the backbone network of YOLOv5s is replaced by a lightweight network, MobilenetV3, to reduce the number of parameters and computation of the model, and the SPPF module of the model is retained to improve the model's ability to detect targets of different sizes; secondly, an attention mechanism is added between the neck and head of the model to better capture the target information in the image and to improve the accuracy and robustness; finally, replace the model loss function with EIoU to accelerate the model convergence and improve the model detection accuracy. The proposed model is trained and validated by the self-constructed helmet dataset, and the experimental results show that compared with the original model, the detection accuracy is improved by 1.2%, the number of parameters is reduced by 39.4%, the computation amount is reduced by 56.3%, and the model volume is compressed by 38.6%, which provides an effective method for the deployment of helmet recognition algorithms based on the improved YOLOv5s on embedded devices.

河北省教育厅科技项目(ZD2022163);中央高校基本科研业务费(3142020048);国家重点研发计划项目(2020YFC1511805)

黎冠,李志伟,陈浩,等.基于改进YOLOv5s的轻量化安全帽佩戴检测算法[J].华北科技学院学报,2024,21(3):32-41

LI Guan, LI Zhiwei, CHEN Hao, et al. Lightweight helmet wearing detection algorithm based on improved YOLOv5s[J].Journal of North China Institute of Science and Technology,2024,21(3):32-41