基于改进YOLOv5的电力故障检测方法  

Power fault detection method based on improved YOLOv5

作　　者：饶梦程 王娜[1,2] 丁军航 叶昱清 RAO Mengcheng;WANG Na;DING Junhang;YE Yuqing(School of Automation,Qingdao University,Qingdao 266071,China;Shandong Provincial Key Laboratory of Industrial Control Technology,Qingdao 266071,China;School of Rehabilitation Sciences and Engineering,University of Health and Rehabilitation Sciences,Qingdao 266071,China)

机构地区：[1]青岛大学自动化学院,山东青岛266071 [2]山东省工业控制技术重点实验室,山东青岛266071 [3]康复大学(筹)康复科学与工程学院,山东青岛266071

出　　处：《电子设计工程》2024年第18期27-31,共5页Electronic Design Engineering

基　　金：2020年山东省自然科学基金重点项目(ZR2020KF034)。

摘　　要：针对无人机电力巡检场景下,线路故障检测的识别速度和识别精度低的问题,提出了一种改进YOLOv5的电力故障检测方法。结合YOLOv5算法基础,在颈部网络中采用轻量化模块,优化了网络结构,减少了网络参数和模型计算的复杂度;替换原模型中的损失函数,进一步提高模型的检测准确性。通过在骨干网络中加入CA注意力机制,使得模型能更准确地识别目标。结果表明,优化后的算法在故障检测数据集上的平均精度均值达到74.8%,相较于YOLOv5模型的平均精度均值提高了4.6%,模型参数量降低了4.2%,检测速度提高了5.3 FPS,总体效果有了明显提升。For the issue of low recognition speed and accuracy in identifying line faults in the scenario of unmanned aerial vehicle power inspection,this article proposes an improved electric fault detection method for YOLOv5.Combined with the basis of the YOLOv5 algorithm,a lightweight module is used in the neck network to optimize the network structure,reduce the complexity of network parameters and model calculations;Replace the loss function in the original model to further improve the detection accuracy of the model.The backbone network is now augmented with the CA attention mechanism,allowing the model to more precisely pinpoint the target.The results showed that the optimized algorithm achieved an average precision mean of 74.8%on the fault detection dataset,which is a 4.6%improvement compared to the original YOLOv5 model.Additionally,the optimized algorithm reduced model parameters by 4.2%and increased detection speed by 5.3 FPS,resulting in a significant overall improvement in performance.

关 键 词：YOLOv5 SIOU损失函数 轻量化模型 注意力机制 

分 类 号：TN919.8[电子电信—通信与信息系统]