基于改进YOLOv5的电厂人员吸烟检测

Smoking detection of power plant personnel based on improved YOLOv5

摘要:

发电厂厂区内违规吸烟易导致火灾、爆炸等事故，会带来巨大损失。针对电厂内人员违规吸烟行为检测精度不高的问题，提出一种基于改进YOLOv5s（You Only Look Once v5s）的电厂内人员违规吸烟检测方法。该方法以YOLOv5s网络为基础，将YOLOv5s网络C3模块Bottleneck中的3×3卷积替换为多头自注意力层以提高算法的学习能力；接着在网络中添加ECA（Efficient Channel Attention）注意力模块，让网络更加关注待检测目标；同时将YOLOv5s网络的损失函数替换为SIoU（Scylla Intersection over Union），进一步提高算法的检测精度；最后采用加权双向特征金字塔网络（BiFPN，Bidirectional Feature Pyramid Network）代替原先YOLOv5s的特征金字塔网络，快速进行多尺度特征融合。实验结果表明，改进后算法吸烟行为的检测精度为89.3%，与改进前算法相比平均精度均值（mAP，mean Average Precision）提高了2.2%，检测效果显著提升，具有较高应用价值。

Abstract:

For power plants, smoking in the factory area is easy to cause fire, explosion and other safety hazards, and improper handling will bring huge losses. Aiming at the problem that the detection accuracy of illegal smoking behavior of personnel in power plant is not high, this paper proposed an improved YOLOv5s (You Only Look Once V5S) target detection method to detect illegal smoking behavior of personnel in power plant. Based on the object detection algorithm YOLOv5s, the 3×3 convolution in the C3 module of YOLOv5s network is replaced by a Bottleneck self-attention layer to improve the learning ability of the algorithm. The Efficient Channel Attention module (ECA) is added to the network to make the network pay more Attention to the target to be detected. At the same time, the loss function of YOLOv5s algorithm was changed to Scylla Intersection over Union (SIoU) to further improve the detection accuracy of the algorithm. Finally, the weighted Bidirectional Feature Pyramid Network (BiFPN) is used to replace the original YOLOv5s Feature Pyramid Network to rapidly perform multi-scale Feature fusion. The experimental results show that, compared with the traditional YOLOv5s algorithm, the improved algorithm improves the detection accuracy of smoking behavior, and the mAP (mean Average Precision) is increased by 2.2%. The recognition effect is significantly improved, which proves the effectiveness of the new algorithm.