基于改进YOLOv11的绝缘子缺陷检测算法

Insulator defect detection algorithm based on improved YOLOv11

摘要:

针对现有电力行业绝缘子缺陷检测方法存在计算量大、检测类型单一和检测精度低等问题，提出一种新的绝缘子缺陷检测算法。以YOLOv11为基础模型，在主干网络末端和C2PSA模块中引入BRA注意力机制，增强模型的抗干扰能力和特征提取能力；用动态卷积模块DynamicConv代替C3K2模块中的普通卷积，进一步提高模型对绝缘子缺陷小目标的检测能力；在回归损失计算中使用WIoU损失函数代替原来的CIoU损失函数，增强模型的定位性能和泛化能力，提高检测速度。实验结果表明，改进后的算法对绝缘子及各类缺陷检测的mAP0.5和mAP0.5-0.95分别达到84.8%和62.6%。

Abstract:

Aiming at the problems of large amount of calculation, single detection type and low detection accuracy in the existing insulator defect detection methods in power industry, a new insulator defect detection algorithm is proposed. Based on YOLOv11 model, Bi-level Routing Attention (BRA) attention mechanism is introduced into the terminal of backbone network and C2PSA module to enhance the anti-interference ability and feature extraction ability of the model. Dynamic convolution (DynamicConv) is used to replace ordinary convolution in C3K2 module, which further improves the detection ability of the model for small targets of insulator defects. In regression loss calculation, WIoU loss function is used to replace the original CIoU loss function, which enhances the positioning performance and generalization ability of the model and improves the detection speed. The experimental results show that mAP0.5 and mAP0.5-0.95 of the improved algorithm for insulator and various defects detection reach 84.8% and 62.6% respectively.