Computer Engineering and Applications ›› 2024, Vol. 60 ›› Issue (21): 172-182.DOI: 10.3778/j.issn.1002-8331.2402-0147

• Graphics and Image Processing • Previous Articles     Next Articles

Safety Helmet Detection Method in Complex Environment Based on Multi-Mechanism Optimization of YOLOv8

XIAO Zhenjiu, YAN Su, QU Haicheng   

  1. School of Software, Liaoning Technical University, Huludao, Liaoning 125105, China
  • Online:2024-11-01 Published:2024-10-25

基于多重机制优化YOLOv8的复杂环境下安全帽检测方法

肖振久,严肃,曲海成   

  1. 辽宁工程技术大学 软件学院,辽宁 葫芦岛 125105

Abstract: In order to solve the problem of low detection accuracy of existing helmet detection algorithms for small targets, dense targets and complex environments in construction sites, tunnels, coal mines and other construction scenes, a helmet detection method based on multiple mechanisms is proposed. Firstly, the C2f module of the Backbone part of YOLOv8n is added with a scalable residual (DWR) attention module, which makes the network more flexible to adapt to features of different scales and accurately identify objects in the image. Secondly, the original Conv in the main part is replaced by the deformable convolution AKConv module, which brings significant performance improvement to the convolutional neural network and achieves more efficient feature extraction. In addition, the combination of large separable kernel attention LSKA module and SPPF structure is used to greatly enhance the fusion capability of the model core. The experimental results on the Safety helmet dataset show that compared with the original model, the improved algorithm has improvement of 10.5 percentage points in mAP@0.5 and 3.7 percentage points in mAP@0.5-0.95, which can effectively improve the accuracy of safety helmet wear detection in complex scenes.

Key words: safety helmet, YOLOv8n, dilation-wise residual (DWR), arbitrary kernel convolution (AKConv), large separable kernel attention (LSKA)

摘要: 为了解决建筑工地、隧道、煤矿等施工场景中现有安全帽检测算法对于小目标、密集目标以及复杂环境下的检测精度低的问题,提出了一种基于多重机制的安全帽检测方法。以YOLOv8n为基础将Backbone部分的C2f模块加入可扩张残差(DWR)注意力模块,使得网络能够更灵活地适应不同尺度的特征,以而更准确地识别图像中的物体;采用可变形卷积AKConv模块取代主干部分中的原始Conv,为卷积神经网络带来了显著的性能提升,从而实现更高效的特征提取。此外引用了大型可分离核注意力LSKA模块与SPPF结构相结合,大大增强了模型核心的融合能力。在Safety helmet数据集的实验结果表明,改进后的算法相较于原模型,mAP@0.5指标上提升了10.5个百分点,在mAP@0.5-0.95指标上提升了3.7个百分点,能有效提高复杂场景下的安全帽佩戴检测精度。

关键词: 安全帽, YOLOv8n, DWR模块, AKConv模块, LSKA模块