摘要
现有的检测方法在复杂背景的输电线路图像中识别绝缘子微小缺陷时,得到的图像存在背景环境复杂、缺陷尺寸小等问题。为保证输电线路的安全运行,提出一种基于YOLOv7-tiny的绝缘子缺陷检测网络(IDD-Net)。首先,引入基于注意力的尺度内特征交互(AIFI)来处理高维特征,从而降低计算量;其次,使用双向加权路径特征金字塔网络(BiFPN)进行特征融合,并对下采样模块进行改进,增强网络的感知能力;最后,使用Focal-DIoU损失函数提高锚框质量。结果表明,与基线模型相比,IDD-Net的平均精度均值提高4.1%,精确率和召回率分别提高2.4%和6.5%,参数量和浮点运算量分别减少5.8%和2.3%,对于闪络缺陷的平均精度提高11.2%。由此说明所提方法参数量较小,性能更优异,鲁棒性更强。
When the existing detection methods identify small defects of insulators in transmission line images with complex backgrounds,the obtained images have problems such as complex background environment and small defect size.In order to ensure the safe operation of transmission lines,an insulator defect detection network(IDD-Net)based on YOLOv7-tiny is proposed.The attention-based intra-scale feature interaction(AIFI)is introduced to handle high-dimensional features and reduce computational complexity.The bidirectional weighted path feature pyramid network(BPFPN)is used for the feature fusion.The improvements are made to the down-sampling module,so as to enhance the network′s perceptual capabilities.The Focal-DIoU loss function is used to improve anchor box quality.The results show that,in comparison with the baseline model,the average accuracy of IDD-Net is improved by 4.1%,the accuracy and recall are improved by 2.4%and 6.5%,the number of parameters and floating-point operations are reduced by 5.8%and 2.3%,respectively,and the average accuracy for flashover defects is improved by 11.2%.It demonstrates that the proposed method has smaller parameters,better performance and stronger robustness.
作者
韩兴宇
陈为真
HAN Xingyu;CHEN Weizhen(School of Electrical and Electronic Engineering,Wuhan Polytechnic University,Wuhan 430048,China)
出处
《现代电子技术》
北大核心
2025年第16期105-112,共8页
Modern Electronics Technique
基金
湖北省教育厅科技项目(B2020061)
湖北省自然科学基金项目(2022CFB449)。
作者简介
韩兴宇(1999-),男,湖北襄阳人,硕士研究生,研究方向为深度学习与目标检测;陈为真(1976-),女,湖北竹山人,博士研究生,副教授,研究方向为智能预测建模及应用、智能计算与决策。
