An insulator self-blast detection method based on YOLOv4 with aerial images

Energy Reports

Xile Huang¹  Yuxuan Song²  Zheng Zhang²  Bo Chen²  Meng Wang²  Hui He²  Guangwei Yan³ 

[1] Corresponding author.;School of Control and Computer Engineering, North China Electric Power University, Beijing 102206, China;Technology Center, Taikang Instance Group Inc., Beijing 102206, PR China;

关键词: Insulator self-blast;    YOLOv4;    Feature fusion;    Attention mechanism;    SENet;    Aerial images;   

DOI  :  

来源: DOAJ

【 摘 要 】

Due to long-term exposure to the natural environment, insulators are prone to self-blast, threatening the safety and reliability of transmission lines. Because of the different sizes of the insulator self-blast area and the complicated background, it is inevitable that the missed and false detections occur. To solve this problem, this paper proposes a deep neural network called Mina-Net (Multi-Layer INformation Fusion and Attention Mechanism Network). Mina-Net is based on YOLOv4. First, the shallow feature map with more detailed texture information is fused into the feature pyramid. Then, an improved SENet is applied to recalibrate the features of different levels in the channel direction. The experimental results on the actual dataset show that Mina-Net has increased the average precision by 4.78% compared with the YOLOv4.

【 授权许可】

Unknown