摘要
针对变电站绝缘套管过热红外图像检测精度不高的问题,提出了基于改进YOLO第7版(you only look once version 7,YOLOv7)算法的检测技术。通过引入改良的跨阶段部分网络幽灵版本3(cross stage partial network ghost version 3,C3Ghost)模块替换头部网络中的扩展高效层聚合网络(extended efficient layer aggregation network,E-ELAN)模块,优化了网络结构,增强了算法对小目标的识别能力。此外,整合了轻量级基于归一化的注意力模块(normalization-based attention module,NAM)到主干网络中以提高对红外图像特征的利用效率,并引入幽灵卷积(ghost convolution,GhostConv)模块替换了网络中的所有卷积,显著降低了模型的大小。结果表明,与YOLOv7初始算法相比,改进YOLOv7算法在F1评分和平均精确率均值上分别提高了19.51%和16.57%,算法的参数量减小了16.3 MB,且检测速度达到了41帧/s,充分证明了该算法在变电站实际应用中的有效性。该研究不仅显著提高了变电站绝缘套管过热红外图像检测的准确性,也能为后续相关技术的研究提供参考。
To address the issue of insufficient accuracy in the detection of infrared images of overheated insulators in substations,a detection technology based on the improved you only look once version 7(YOLOv7)algorithm was proposed.The network structure was optimized by introducing an improved cross stage partial network ghost version 3(C3Ghost)module to replace the extended efficient layer aggregation network(E-ELAN)module in the head network,thereby enhancing the model′s ability to recognize small targets.Additionally,a lightweight normalization-based attention module(NAM)was integrated into the backbone network to improve the efficiency of infrared image feature utilization.Furthermore,all convolutions in the network were replaced with ghost convolution(GhostConv)modules,significantly reducing the model size.The results indicated that,compared to the original YOLOv7 algorithm,the improved YOLOv7 algorithm increased the F1 score and mean average precision by 19.51%and 16.57%,respectively,while reducing the model parameters by 16.3 MB and achieving a detection speed of 41 frame/s,which fully demonstrated its effectiveness in practical substation applications.This research not only significantly improved the infrared image detection accuracy of overheated insulation bushing in substations but also provided a reference for subsequent studies in related technologies.
作者
肖天龙
何昕怡
李云
朱黎
XIAO Tianlong;HE Xinyi;LI Yun;ZHU Li(College of Intelligent Systems Science and Engineering,Hubei Minzu University,Enshi 445000,China)
出处
《湖北民族大学学报（自然科学版）》
CAS
2024年第3期349-354,共6页
Journal of Hubei Minzu University:Natural Science Edition
基金
国家自然科学基金项目(61961017)。
