摘要
国内的电力事业发展迅速,输电杆塔的缺陷检测与修复是保证电网安全运行的关键技术手段。当前主要是人为识别输电杆塔的缺陷,工作负担巨大。故以YOLOv5s网络为基础,提出一种改进YOLOv5s目标检测算法,提升检测效率。在基础模型上引入Focal-EIoU损失函数,提升模型收敛速度与精度;在卷积层引入Hardswish激活函数,提高模型的表达能力,查准率得到提升;上调算法推理的置信度阈值conf-thres,减少模型推理的误检情况,提升模型正检率。另外在研究中尝试融入注意力机制提升网络特征提取能力,但效果不好,故舍弃此改进策略。实验结果表明,改进模型的各项指标均获得了提升,查准率由92.96%提升至95.02%,上涨了2.06百分点;查全率由87.36%提升到了87.38%;mAP@.5∶mAP@.5∶.95(0.1∶0.9)由0.644 3提升至0.648 1,上涨了0.38百分点;模型检测速度FPS提高了4.4。
With the rapid development of domestic electric power industry,the defect detection and repair of transmission towers are key technical means to ensure the safe operation of the power grid.At present,it is mainly to identify the defects of transmission tower manually,and the work burden is huge.Therefore,we propose an improved YOLOv5s object detection algorithm based on the YOLOv5s network to improve the efficiency of detection.Focal-EIoU loss function is introduced into the basic model to improve the convergence speed and accuracy of the model.The Hardswish activation function is introduced into the convolution layer to improve the expression ability of the model and the precision.The confidence threshold conf-thres of the algorithm reasoning is increased to reduce the false detection of model reasoning and improve the positive detection rate of the model.In addition,in the research,we tried to integrate attention mechanism to improve the ability of network feature extraction,but the effect was not ideal,so we abandoned this improvement strategy.The experimental results show that all indicators of the improved model have been improved,with a precision increase of 2.06 percent from 92.96%to 95.02%;the recall rate has increased from 87.36%to 87.38%;mAP@.5∶mAP@.5∶.95(0.1∶0.9)increased from 0.6443 to 0.6481,an increase of 0.38 percent;the model detection speed FPS has been improved by 4.4.
作者
冀承泽
贾立新
李荆晖
JI Cheng-ze;JIA Li-xin;LI Jing-hui(School of Electrical Engineering,Xi'an Jiaotong University,Xi'an 710049,China)
出处
《计算机技术与发展》
2024年第2期180-185,共6页
Computer Technology and Development
