无线电工程

针对目前绝缘子缺陷目标检测算法中存在的误检、漏检和检测精度低等问题，提出一种基于改进YOLOv8的高压输电线路绝缘子缺陷检测方法，实现了高精度检测。在改进的YOLOv8模型中，基于可变形卷积神经网络(Deformable Convolutional Neural Network, DCNN)和全局注意力机制(Global Attention Mechanism, GAM)设计了可变形注意力骨干网络，减少了特征提取过程中有效目标特征的丢失；基于卷积块注意力模块(Convolutional Block Attention Module, CBAM),提出改进的空间金字塔池化快速特征融合(Spatial Pyramid Pooling Fast Feature Fusion, SPFF)模块，结合高效通道注意力(Efficient Channel Attention, ECA)机制，扩大了模型的感受野，保留了更多类型的绝缘子缺陷特征信息，提高了检测精度；采用稳定交并比(Stable Intersection over Union, SIoU)损失函数，加快了模型的收敛速度，提升了对小目标缺陷的检测能力；构建了一个包含“Normal”“Defect”“Broke”“Flashover”四种类型的绝缘子缺陷数据集。实验结果表明，改进后的YOLOv8模型的平均精度均值(mean Average Precision, mAP)达到95.84%,较原YOLOv8提高了5.58%,在各类绝缘子上的AP值均显著优于其他算法。相比原始算法，改进后的YOLOv8模型在小目标缺陷检测方面的表现显著提升，进一步验证了所提算法在绝缘子缺陷检测中的可行性和有效性。

To address the issues of false detection, missed detection and low detection accuracy in existing insulator defect detection algorithms, a high-voltage transmission line insulator defect detection method based on improved YOLOv8 model is proposed for high-accuracy detection. First, a deformable attention backbone network is designed using a Deformable Convolutional Neural Network(DCNN) combined with a Global Attention Mechanism(GAM), reducing the loss of effective object features during feature extraction. Next, an improved Spatial Pyramid Pooling Fast Feature Fusion(SPFF) module is proposed based on the Convolutional Block Attention Module(CBAM), combining with the Efficient Channel Attention(ECA) mechanism to expand the model's receptive field and preserve more types of defect features, thereby improving the detection accuracy. Additionally, the Stable Intersection over Union(SIoU) loss function is introduced to speed up the model's convergence and enhance small object defect detection capabilities. Finally, a dataset containing four types of insulator defects, i.e. Normal, Defect, Broke, Flashover, is constructed. Experimental results show that the improved YOLOv8 model achieves an mean Average Precision(mAP) of 95.84%, which is 5.58% higher than the original YOLOv8, and outperforms other models in terms of AP values across all types of insulators. The improved YOLOv8 model also shows significant improvement in detecting small object defects compared to the original algorithm, further validating its feasibility and effectiveness for insulator defect detection.