重庆科技大学学报（自然科学版）2026，Vol.28Issue(1)：60-69,111,11.DOI:10.19406/j.issn.2097-4531.2026.01.007
基于改进YOLOv8的接触网异物检测模型研究
Research on Foreign Object Detection Model for Catenary Based on Improved YOLOv8
摘要
Abstract
To address the problems of insufficient feature extraction,severe background interference,and low detec-tion accuracy in existing catenary foreign object detection methods,we propose an improved YOLOv8-based detec-tion model.First,a C2f-SCConv module is introduced into the feature extraction backbone of YOLOv8 to enhance the model's adaptability to foreign object features and background variations.Then,we embed a BiFPN structure to effectively fuse multi-scale feature information,thereby improving detection accuracy for objects of different scales.Furthermore,a GAM mechanism is incorporated to strengthen image noise suppression and enhance model robust-ness.Finally,the Wise-IoU loss function is introduced to increase sensitivity to complex object boundaries and en-hance fine-grained feature capabilities.Experimental results demonstrate that the improved model achieves a mean Average Precision of 89.9％in catenary foreign object detection,significantly improving the detection performance for small targets and meeting the practical requirements of catenary foreign object inspection.关键词
接触网/异物检测/YOLOv8模型/SCConv模块/铁路安全Key words
Catenary/foreign object detection/YOLOv8 model/SCConv module/railway safety分类
信息技术与安全科学引用本文复制引用
李作进,郑路,徐椤庚,晋智炜,李自力,万久地..基于改进YOLOv8的接触网异物检测模型研究[J].重庆科技大学学报（自然科学版）,2026,28(1):60-69,111,11.基金项目
重庆市自然科学基金项目"面向多模态异构大数据的特征自主学习方法研究"(CSTC2021YCJH-BGZXM0071) （CSTC2021YCJH-BGZXM0071）
重庆市教委科技重大项目"山地道路疲劳驾驶特征融合与险态行为识别研究"(KJZD-M202301502) （KJZD-M202301502）
