长沙理工大学学报（自然科学版）2025，Vol.22Issue(2)：120-131,12.DOI:10.19951/j.cnki.1672-9331.20241103001
基于轻量级SGG-YOLOv8n的X-DR电缆开关故障诊断
Fault diagnosis of X-DR cable switch based on lightweight SGG-YOLOv8n
摘要
Abstract
[Purposes]In view of problems such as complex model and low detection efficiency in cable switch image detection algorithm,this paper explored an improved lightweight YOLOv8n model based on the global attention mechanism(GAM)and ghost convolution(GhostConv),namely SGG-YOLOv8n,so as to achieve rapid and accurate diagnosis of cable switch faults in X-ray digital radiography(X-DR).[Methods]Based on the YOLOv8n model,spatial and channel reconstruction convolution(SCConv)was introduced into the backbone network and neck network and replaced the Bottleneck of the C2f module,so as to make the model's receptive field more flexible and improve the detection efficiency and performance of the model.At the same time,GhostConv was introduced into the neck network of the YOLOv8n model to replace the original Conv module,reducing the amount of model computation and number of parameters and ensuring a lightweight model.Moreover,GAM was introduced in front of the detection head to enhance the model's ability to perceive the global feature information.[Findings]Experiments on the X-DR cable switch image dataset show that compared with YOLOv8n,the SGG-YOLOv8n model proposed in this paper reduces the number of parameters by 0.39×106,decreases the floating-point operations per second by 0.9,and improves the precision and recall rate by 1.4 percentage points.In addition,the model raises the mean average precision at the intersection over union(IoU)thresholds of 0.5 and 0.5-0.95 by 1.7 percentage points and 1.2 percentage points,respectively.The model processing speed reaches 65 frames per second.[Conclusions]The proposed method has been applied to the fault diagnosis of the X-DR cable switch,and the fault diagnosis rate for cable switches has been improved from 87.9%to 91.4%after the SGG-YOLOv8n model is applied.关键词
电缆开关/X射线数字成像/YOLOv8n/空间与通道重建卷积/幻影卷积/注意力机制Key words
cable switch/X-ray digital radiography/YOLOv8n/spatial and channel reconstruction convolution/ghost convolution/attention mechanism分类
动力与电气工程引用本文复制引用
秦学,李爱平,王长河,樊军,程宏波,朱建勇,刘金平..基于轻量级SGG-YOLOv8n的X-DR电缆开关故障诊断[J].长沙理工大学学报（自然科学版）,2025,22(2):120-131,12.基金项目
国家自然科学基金资助项目(62363010) （62363010）
江西省主要学科学术带头人培养项目(20232BCJ22004) Project(62363010)supported by the National Natural Science Foundation of China （20232BCJ22004）
Project of Cultivating Academic Leaders in Key Disciplines of Jiangxi Province(20232BCJ22004) （20232BCJ22004）
