摘 要：文章提出了一种基于 Scaled-YOLOv4 目标检测方法的破损绝缘子智能检测模型。针对 Scaled-YOLOv4 网络在训练过程中难以分辨有效信息的问题，分析 Scaled-YOLOv4 网络 Neck 部分的降采样操作会导致信息丢失，提出将改进的注意力机制加入网络模型中，设计了 DC-Scaled-YOLOv4 模型。将网络上得到的破损绝缘子数据集分配成训练集和测试集，并对故障识别模型进行训练。采用该模型对破损绝缘子进行识别测试，Scaled-YOLOv4 在破损绝缘子数据集上的检测精度为 80%，而文章算法在破损绝缘子数据集上的检测精度为 94.8%，检测效果提升明显。
关键词：目标检测；Scaled-YOLOv4；注意力机制；绝缘子
DOI:10.19850/j.cnki.2096-4706.2022.04.032
中图分类号：TP393 文献标识码：A 文章编号：2096-4706（2022）04-0123-04
Research on Damaged Insulator Detection Based on Deep Learning
WANG Ying, WU Jiansheng
(School of Computer Science and Software Engineering, University of Science and Technology Liaoning, Anshan 114051, China)
Abstract: This paper proposes an intelligent detection model of damaged insulator based on Scaled-YOLOv4 target detection method. Aiming at the problem that it is difficult for Scaled-YOLOv4 network to distinguish effective information in the training process, it is analyzed that the down sampling operation of Neck part of Scaled-YOLOv4 network will lead to information loss. Adding an improved attention mechanism to the network model is proposed, and a DC-Scaled-YOLOv4 model is designed. The damaged insulator data set obtained from network is allocated into training set and test set, and the fault identification model is trained. The model is used to identify and test the damaged insulator. The detection accuracy of Scaled-YOLOv4 on the damaged insulator data set is 80%, while the detection accuracy of algorithm proposed in this paper on the damaged insulator data set is 94.8%, and the detection effect is significantly improved.
Keywords: target detection; Scaled-YOLOv4; attention mechanism; insulator
参考文献：
[1] WOO S，PARK J，LEE J Y，et al.CBAM： Convolutional Block Attention Module [J/OL].arXiv：1807.06521[cs.CV].[2022-01-03]. https://arxiv.org/abs/1807.06521.
[2] CHOLLET F.Xception：Deep Learning with Depthwise Separable Convolutions [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Hawaii：IEEE，2017：1800-1807.
[3] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：
Towards Real-Time Object Detection with Region Proposal Networks [J].IEEE Transactions on Pattern Analysis and Machine Intelligence， 2017，39（6）：1137-1149.
[4] WANG C Y，BOCHKOVSKIY A，LIAO H Y M.ScaledYOLOv4：Scaling Cross Stage Partial Network [J/OL].arXiv： 2011.08036[cs.CV].[2022-01-02].https://arxiv.org/abs/2011.08036.
[5] LIN T Y，DOLLAR P，GIRSHICK R, et al. Feature Pyramid Networks for Object Detection [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Honolulu：IEEE，2017：936-944.
作者简介：王迎（1997.11—），女，汉族，辽宁锦州人，硕士研究生，研究方向：计算机视觉。