N-YOLOv8:绝缘子缺陷自动检测模型OA

Objective The normal operation of insulators is a crucial part of ensuring the safe and reliable operation of power systems.To promote the automatic detection of insulator defects,a lightweight detection model,N-YOLOv8,was designed.Methods For network lightweighting,a lightweight module called DPHConv was first designed by integrating depthwise separable convolution,BN layer,Hardswish activation function,and residual connections.Second,a lightweight module,CFI-X,was designed by integrating Inception-Bottleneck and C2f.These two modules significantly reduced the number of network parameters at the cost of only a minor loss in detection accuracy.In terms of improving detection accuracy,a T-ECA module was designed based on the ECA attention mechanism.Then,an attention mechanism PEG was designed by integrating the T-ECA module with a multi-branch parallel structure and a Gather Excite attention module.Furthermore,CEIoU_Loss was designed by combining the ideas of CIoU_Loss and EIoU_Loss.Finally,Soft-NMS was introduced to replace the NMS in the original network.These improvements effectively boosted the network's performance in detecting insulator defects.Results Compared with the YOLOv8n network,N-YOLOv8 reduced the number of parameters by 43％and floating-point operations by 37％,while achieving a detection accuracy of 91.7％,which was 0.2％higher than that of the original network.Conclusion N-YOLOv8 delivers high detection performance,which can effectively facilitate the deployment of intelligent detection algorithms on UAV platforms and achieve real-time detection of insulator defects.