基于YOLOv8s的轻量级绝缘子多缺陷检测模型

Lightweight insulator multi-defect detection model based on YOLOv8s

YOLO系列算法已广泛用于识别电力线路中的各类缺陷目标。由于巡检图像背景复杂、缺陷目标的尺度不一等,直接利用YOLO算法难以有效避免绝缘子闪络、破损等小目标的错检漏检问题。为解决这一问题,在YOLOv8s模型的基础上提出一种轻量化绝缘子缺陷检测算法。在骨干网络中引入双层路由注意力机制(BRA),以提升对全局特征的关注度,抑制背景噪声,降低小目标缺陷的错检漏检率。通过加权双向特征金字塔网络(BiFPN)实现跨尺度特征之间的加权融合,获取各类缺陷更全面的特征信息。重构Neck网络来消除低贡献度的网络节点,在增强检测性能的同时减少了模型的参数量,实现了性能提升和参数效率之间的平衡。实验结果显示,改进后的网络模型平均检测精度达到84.9%,而参数量仅为8.4×10^(6),可实现对绝缘子缺陷的快速准确检测。

The YOLO series algorithm has been widely used to identify various types of defect targets in power lines.Due to the complex background of inspection images and the varying sizes of defect targets,it is difficult to effectively avoid the problem of false detection and missed detection of small targets such as insulator flashover and damage by means of the YOLO algorithm.On this basis,a lightweight insulator defect detection algorithm is proposed on the basis of the YOLOv8s model.The bi-level routing attention(BRA)mechanism is inserted into the backbone network to enhance attention to global features,suppress background noise,and reduce the false positive detection rate of small target defects.The weighted bidirectional feature pyramid network(BiFPN)is used to achieve the weighted fusion of cross-scale features and obtain more comprehensive feature information of various defects.The Neck network is reconstructed to eliminate the low-contribution network nodes,and the number of parameters is reduced while the detection performance is enhanced,thus achieving a balance between performance improvement and parameter efficiency.The experimental results show that the improved network model has an average detection accuracy of 84.9%and a parameter size of only 8.4×10^(6),which can achieve fast and accurate detection of insulator defects.