基于通道注意力的多任务卡具缺损检测算法  

Multi⁃task fixture defect detection algorithm based on channel attention

作　　者：赵佩仪 ZHAO Peiyi(Pittsburgh Institute,Sichuan University,Chengdu 610207,China)

机构地区：[1]四川大学匹兹堡学院,四川成都610207

出　　处：《电子设计工程》2025年第1期190-195,共6页Electronic Design Engineering

摘　　要：为及时地发现铁路通信漏缆卡具缺损现象,保障铁路运营安全,该文提出一种基于通道注意力的多任务卡具缺损检测算法。在YOLOv7-tiny的基础上,该算法通过在检测头增加通道注意力机制的方式,平衡不同通道之间的重要程度,提高了模型检测能力。为了实现对卡具的状态检测,算法在类别损失函数中额外增加一项状态损失,并在其中嵌入Focal loss来克服样本不足的问题。为了验证算法的有效性,该文分别对实际线路中的四种类型卡具进行实验。实验结果表明,该文提出的算法效果好、速度快,能满足实际现场需求。To discover the broken fixtures of railway communication cable in time and guarantee the safety operation for the railway,this paper proposes a multi⁃task fixture defect detection algorithm based on channel attention.Based on YOLOv7⁃tiny,we embed a Channel Attention Module(CAM)to balance the importance of different channels.In order to achieve state detection of fixtures,we add an additional state loss in the class loss function and embed Focal loss to overcome imbalance problem.To verify the effectiveness of the algorithm,experiments were conducted on real⁃world dataset containing four kinds of fixtures.The results have shown that our algorithm has a high performance and rapid speed,which can meet practical requirements.

关 键 词：铁路巡检 卡具检测 深度学习 通道注意力 YOLOv7 

分 类 号：TN99[电子电信—信号与信息处理]