【摘要】 由于无人机电力影像存在绝缘子器件尺度变化大，输电线路背景复杂，绝缘子缺陷目标小的特点，导致传统目标检测算法识别精度不高。该文提出以YOLOv5l为基础的CA、Transformer编码块和多尺度相融合的改进网络模型，较好的提高了大尺度变化影像上绝缘子缺陷检测的准确性，提升了复杂背景下多类型绝缘子缺陷识别的能力，并解决了微小绝缘子缺陷漏检的问题。基于在某电网公司的数据集上完成训练和验证实验，表明优化模型相比原YOLOv5l模型，准确率提升8.9%,召回率提升4.4%,平均精度均值提升3.5%,说明改进模型对绝缘子缺陷检测有效。更多还原

【Abstract】 Due to the characteristics of UAV power image, such as large variation of insulator device scale, complex transmission line background and small insulator defect target, the recognition accuracy of traditional target detection algorithm is not high. This paper proposes an improved network model based on YOLOv5l, CA, Transformer coding block and multi-scale integration, which can better improve the accuracy of insulator defect detection on large-scale change images, improve the ability of multi-type insulator defect identification under complex background, and solve the problem of small insulator defect detection.Extensive training and verification experiments were conducted on a comprehensive dataset provided by a power grid company. The experimental results demonstrate that the optimized model achieves a significant enhancement over the original YOLOv5l model, with an increase of 8.9% in accuracy rate, 4.4% in recall rate, and 3.5% in average accuracy. Moreover, comparative analysis with the state-of-the-art detection model confirms the effectiveness and reliability of our improved model for insulator defect detection.更多还原