山东电力技术2025，Vol.52Issue(5)：18-29,12.DOI:10.20097/j.cnki.issn1007-9904.2025.05.003
电力钢绞线多尺度损伤图像检测方法
Multi-scale Damage Detection Method for Power Steel Strand Image
摘要
Abstract
To address the challenges in detecting steel strand damage in high-voltage overhead lines,an enhanced YOLOv7 model for detecting steel strand damage is proposed.Given the issue of small-scale damage during the detection process,a Coordinate Attention module is employed to enhance the network's ability to extract information from small-scale targets.Based on the pyramid feature fusion architecture of the path aggregation feature pyramid network(PaFPN),an adaptively spatial feature fusion(ASFF)module is added to enhance the ability of the module to handle images of different sizes and various damage scales.Additionally,a weighted intersection over union is used to replace the complete intersection over union in optimizing the loss function,and a dynamic focusing mechanism is applied to bounding box regression to enhance the module's robustness.Experimental results demonstrate a significant improvement in the performance of the modified YOLOv7 module compared to the original YOLOv7 module in detecting steel strand damage,with a 15.8%increase in mean average precision.The proposed model outperforms the original model in the effectiveness of detecting steel strand damage.关键词
目标检测/钢绞线/多尺度损伤/YOLOv7/注意力机制/损失函数/自适应机制Key words
object detection/steel strand/multi-scale damage/YOLOv7/attention mechanism/loss function/adaptive mechanism分类
计算机与自动化引用本文复制引用
陈宇,缪进荣,杨辰飙,颜钰霆,贺润平..电力钢绞线多尺度损伤图像检测方法[J].山东电力技术,2025,52(5):18-29,12.基金项目
国家电网有限公司科技项目(520935220004).Science and Technology Project of State Grid Corporation of China(520935220004). （520935220004）
