基于改进YOLOv8的输电线路覆冰检测

Icing Detection of Transmission Lines Based on Improved YOLOv8

中文关键词:  深度学习  注意力机制  YOLOv8  输电线路覆冰检测

英文关键词:deep learning  attention mechanism  YOLOv8  icing detection of transmission lines

中文摘要:

      输电线路覆冰会给输电线路的安全稳定造成严重影响。由于输电线路多分布在山区、林区、无人的空旷地带，发生雨雪冰冻等自然灾害时工作人员无法在第一时间获取现场信息，为了实现对复杂环境下的输电线路覆冰的有效在线监测，提出了一种基于YOLOv8的改进的覆冰检测模型和算法。首先，用幽灵洗牌卷积取代了颈部传统卷积，以减少模型参数，降低计算成本；其次，由于在复杂的背景环境下不能充分捕捉线路覆冰的显著特征，引入了BiFormer注意力机制，该机制有效提高了模型对复杂特征的感知能力，抑制了其余不相关的背景信息；最后，将损失函数替换为Wise-IoUv3，配备合理的梯度分配策略，优先考虑平均质量的样本，从而提高了模型的精确定位能力。实验结果表明，改进模型相比于YOLOv8模型，检测精度提升了7.9%。

英文摘要:

      The icing of transmission lines can have a serious impact on the safety and stability of transmission lines.Because the transmission lines are mostly distributed in mountainous areas,forest areas,and unmanned open areas,when damage occurs such as rain,snow,and freezing,the staff cannot obtain on-site information in a timely manner.In order to achieve effective online monitoring of icing on transmission lines in complex environments,an improved icing detection model and algorithm based on YOLOv8 is proposed.Firstly,ghost shuffle convolution replaces some neck ordinary convolution to reduce model parameters and accelerate model convergence speed.Secondly,due to the inability to fully capture the significant features of line icing in complex background environments,the BiFormer attention mechanism is introduced,which effectively improves the models attention to complex features and suppresses other irrelevant background information.Finally,the loss function is replaced with Wise-IoUv3,equipped with a reasonable gradient allocation strategy,prioritizing samples of average quality,so the models precise localization ability is improved.The experimental results show that the improved model improves detection accuracy by 7.9% compared to the YOLOv8 model.