基于改进YOLOv5的电力设备轻量化检测算法

Lightweight Detection Algorithm of Power Equipment Based on Improved YOLOv5

提出一种轻量化红外目标检测算法MEGI-YOLOv5。该算法基于YOLOv5模型,首先将主干网络替换为轻量化Mobilenet-v3网络,并将颈部网络中的部分CBL结构块替换为倒残差结构的深度可分离卷积、C3模块由普通卷积和GhostConv组合代替,降低模型的参数和计算量;其次在颈部网络中嵌入ECA(Efficient Channel Attention)模块,提高模型通道间信息的注意力,从而提升模型特征提取能力。实验结果表明,该模型相较于YOLOv5模型,参数量减少22%,检测速度提升37%,模型检测精度达到96.42%,能满足变电站设备类别及发热点识别的准确性和实时性要求,为后续能够及时发现变电站设备故障提供保障。

A lightweight infrared target detection algorithm MEGI-YOLOv5 was proposed.The algorithm was based on the YOLOv5 model.Firstly,the backbone network was replaced with the lightweight Mobilenet-v3 network,and part of the CBL structure in the neck network was replaced with the deep separable convolution of the reciprocal residual structure.The C3 module was replaced by the combination of ordinary convolution and GhostConv to reduce the model parameters and calculation amount.Secondly,the Efficient Channel Attention(ECA)module was embedded in the neck network to improve the model's attention to the channel,so as to improve the model's feature extraction ability.The experimental results showed that compared with the YOLOv5 model,the number of parameters of the model was reduced by 22%,the detection speed was increased by 37%,and the detection accuracy of the model could reach 96.42%,which could meet the accuracy and real-time requirements of the identification of substation equipment categories and hotspots,and provide conditions for the subsequent timely detection of substation equipment faults.