摘要
针对现有输电线路绝缘子缺陷检测模型在雨天复杂场景下识别效果差、推理速度慢等问题,在YOLOv7-tiny(you only look once version 7-tiny)的基础上提出了RID-YOLOv7(rain insulator detection-YOLOv7)轻量级雨天场景绝缘子缺陷检测模型。首先探索了坐标注意力机制(coordinate attention,CA)在主干特征提取网络中的最优嵌入位置,提升了模型对目标位置关键特征的提取能力;然后在颈部特征融合网络中引入了幻影混洗卷积(ghost shuffle convolution,GSConv)和幻影混洗跨级部分矢量化旋涡(vortex of vectorized ghost shuffle cross stage partial,VoV-GSCSP),大幅降低了推理时间;最后使用了明智交并比(wise intersection over union,WIoU)优化边界框定位损失函数,提高了模型收敛效率。结果表明,与原始YOLOv7-tiny相比,RID-YOLOv7模型的精确率、召回率和平均精确率均值分别提升了2.41%、5.44%和3.22%,推理速度为88.7帧/s,有效平衡了检测速度和精度。该模型适合对雨天场景下输电线路绝缘子缺陷进行实时检测。
Aiming at the problems of poor recognition effect and slow inference speed of existing transmission line insulator defect detection models for insulators in rainy complex scenes,a rain insulator detection-you only look once version 7(RID-YOLOv7)model which was a lightweight insulator defect detection model in rainy scenes was proposed on the basis of YOLOv7-tiny.Firstly,the optimal embedding position of the coordinate attention(CA)mechanism in the backbone feature extraction network was explored to improve the model′s ability to extract key features of the target position.Secondly,ghost shuffle convolution(GSConv)and vortex of vectorized ghost shuffle cross stage partial(VoV-GSCSP)were introduced into the neck feature fusion network to greatly reduce the inference time.Finally,wise intersection over union(WIoU)was used to optimize the bounding box positioning loss function and improve the convergence efficiency of the model.The results showed that compared with the original YOLOv7-tiny,the precision,recall and mean average precision of the RID-YOLOv7 model were improved by 2.41%,5.44%and 3.22%,respectively.The inference speed reached 88.7 frames/s,which effectively balanced the detection speed and accuracy.The model is more suitable for real-time detection of transmission line insulator defects in rainy scenes.
作者
齐浩宇
谭爱国
梁会军
钟建伟
杨永超
陈文涛
QI Haoyu;TAN Aiguo;LIANG Huijun;ZHONG Jianwei;YANG Yongchao;CHEN Wentao(College of Intelligent Systems Science and Engineering,Hubei Minzu University,Enshi 445000,China;Enshi Power Supply Company,State Grid Hubei Electric Power Co.,Ltd.,Enshi 445000,China)
出处
《湖北民族大学学报（自然科学版）》
CAS
2024年第2期233-240,共8页
Journal of Hubei Minzu University:Natural Science Edition
基金
国家自然科学基金项目(62163013)
湖北省重点研发计划项目(2023BAB120)。
关键词
绝缘子缺陷检测
雨天场景
轻量化
坐标注意力机制
损失函数
深度学习
insulator defect detection
rainy scene
lightweight
coordinate attention mechanism
loss function
deep learning
