基于轻量化改进YOLOv8的绝缘子破损识别方法

Insulator Damage Identification Method Based on Lightweight Improved YOLOv8

针对电力线路绝缘子破损目标检测中图像背景复杂重叠、算法模型计算量大的问题,提出基于轻量化改进YOLOv8的绝缘子破损识别方法。在YOLOv8主干网络中引入高效多尺度注意力机制,采用轻量化卷积模块代替标准卷积模块,采用WIoU损失函数改进原有的损失函数。实验结果表明,基于轻量化改进的YOLOv8与原YOLOv8算法相比,绝缘子破损目标检测的平均精度提高4.35%,模型参数量下降了23.53%,表明该方法在算法精度和轻量化改进方面均有提高,为电力线路巡检在边缘端设备部署该算法提供了可能。

Aiming at the problems of complex overlapping image background and large calculation amount of algorithm model in the detection of damaged insulator of power line,a new method of damaged insulator identification based on lightweight and improved YOLOv8 is proposed.An efficient multi-scale attention mechanism is introduced into the YOLOv8 backbone network,a lightweight convolutional module is used to replace the standard convolutional module,and a WIoU loss function is used to improve the original loss function.The experimental results show that compared with the original YOLOv8 algorithm,the average precision of insulator breakage detection based on the lightweight improvement is increased by 4.35%,and the number of model parameters is decreased by 23.53%,indicating that the algorithm has improved in terms of algorithm accuracy and lightweight improvement,which provides a possibility for the deployment of the algorithm on edge inspection equipment of power lines.