军事知识服务平台
融合全维动态卷积与聚焦调制网络的输电线路防振锤缺陷检测模型
Transmission Line Damper Defect Detection Model Integrating Full-Dimensional Dynamic Convolution and Focus Modulation Network
摘要 在输电线路防振锤缺陷检测中,常常面临缺陷样本不平衡、环境背景复杂以及防振锤尺寸较小等挑战.基于此,提出了一种用于输电线路防振锤缺陷检测的FSO-YOLO模型.在YOLOv8s的主干和颈部结构中引入全维动态卷积模块(omni-dimensional dynamic convolution,ODConv),增强了算法对小目标缺陷特征提取和融合的能力;在模型中加入聚焦调制网络(focal modulatoin network,FMNet),提高了模型聚焦关键区域的识别能力;使用样本加权函数Slideloss取代原分类损失函数,以缓解样本不均衡问题.实验结果表明所提出的FSO-YOLO模型能够有效检测输电线路防振锤缺陷,平均精度均值(mean average precision,mAP)可达95.3%,浮点计算量(floating point operations,GFLOPs)比原模型降低了32%,检测速度(frames per second,FPS)为51帧/s,可以更加快速精准地定位识别缺陷.
Abstract:
In the defect detection of dampers for transmission lines,practitioners often face challenges such as imbalanced defect datasets,complex environmental backgrounds,and the small size of these dampers.To address these issues,this paper proposes a FSO-YOLO model specifically designed for the defect detection of transmission line dampers.The FSO-YOLO model integrates an omni-dimensional dynamic convolution(ODConv)module into the backbone and neck architectures of YOLOv8s to enhance the algorithm′s capability in extracting and fusing features of small-scale defect targets;a focal modulation network(FMNet)is incorporated,and the model′s ability to focus on and recognize key areas is enhanced;furthermore,a sample weighting function named Slideloss is employed to replace the original classification loss function,thereby mitigating the problem of sample imbalance.Experimental results demonstrate that the proposed FSO-YOLO model is effective in detecting defects in transmission line dampers,achieving an average accuracy(mAP)of 95.3%,reducing the floating-point operations(GFLOPs)by 32%compared to the baseline model,and achieving a detection speed of 51 frames per second(FPS),enabling faster and more accurate localization and identification of defective targets.
刊 名 南方电网技术 ISTICPKU
年,卷(期): 2026, 20(9)
分类号: TM755
机标分类号: TP393.08 TN925.93 TN722.75
在线出版日期: 2026年10月27日