基于深度学习的智能电网违章识别算法研究
Research on Smart Grid Violation Recognition Algorithm Based on Deep Learning
靳健欣 1武文起 2张玮 1卢峰超 1田志友 1靳元园1
作者信息
- 1. 河北电力工程监理有限公司,河北 石家庄 050021
- 2. 河北水利电力学院,河北 沧州 061001
- 折叠
摘要
针对智能电网基建施工现场违章识别中安全帽检测在复杂场景下检测精度低的问题,提出了一种基于YOLOv7改进的安全帽佩戴检测方法,首先在模型主干网络中引入MCA,获取更多不同感受野下的特征信息,增强网络对感兴趣区域的关注能力并提升对小目标的识别能力,其次利用检测目标类圆特性,设计使用BCR代替BBR对目标进行学习,以减少图像背景信息干扰,提升密集目标识别率,最后针对检测圆提出基于cIoU的动态焦点Focal-cWIoU损失函数,动态调整几何惩罚项,降低低质量样本的影响,提升模型的检测精度,试验结果表明,该方法的检测时间和精度均满足电力基建现场各种复杂施工场景的要求.
Abstract
Aiming at the problem of low detection accuracy in complex scenes such as dense detection targets and large scale differences in safety helmet detection in smart grid infrastructure construction site violation identification,an improved safety helmet wearing detection method based on YOLOv7 was proposed.Firstly the multi-scale coordinate attention mechanism(MCA)was added to the backbone of model,it improved the ability to recognize the small objects with more feature informa-tion under different receptive field.Secondly,by using the circular characteristics of the detection target,BCR was designed to replace BBR to learn the target,so as to reduce the interference of image background information and improve the recognition rate of dense targets.A dynamic Focal-cWIoU loss function based on cIoU was proposed to dynamically adjust the geometric penalty term,reduce the influence of low-quality samples and improve the detection accuracy of the model.The test results showed that the detection time and accuracy of this method could meet the requirements of various complex construction scenes in power infrastructure site.
关键词
电网基建/安全帽检测/YOLOv7/Focal-cWIoU/多尺度融合坐标注意力机制Key words
power grid infrastructure construction/safety helmet detection/YOLOv7/focal circle wise intersection over union(Focal-cWIoU)/multi-scale coordinate attention mechanism引用本文复制引用
出版年
2024
NETL
NSTL
万方数据