航拍绝缘子自爆缺陷的轻量化检测方法

Lightweight Detection Method of Self-explosion Defect of Aerial Photo Insulator

为了精准识别、定位架空输电线路中航拍绝缘子串的自爆缺陷,提出一种轻量化检测方法MDD-YOLOv3。首先将YOLOv3主干网络残差单元中的普通卷积替换为深度可分离卷积,设计主干网络D-Darknet53,在网络检测精度微降的情况下,大幅提升网络的检测速度。特征挖掘模块中,设计了Dense-SPP模块,Dense-SPP和它前后串联的卷积特征提取层能充分挖掘自爆缺陷的全局和局部特征,提高网络对自爆缺陷的特征表达能力。最后构建了四维度预测层,能充分提取自爆缺陷的位置、纹理和语义等信息,提高网络的小目标检测性能。仿真实验表明,MDD-YOLOv3对绝缘子的检测精确度达到96.1%,检测速度达到36帧/s,相比YOLOv3,检测精确度和速度分别提升了4.0%和28.6%。研究结果证明所提方法可以在复杂背景下快速且精准的识别和定位绝缘子缺陷。

In order to accurately identify and locate the self-explosion defects of aerial insulator strings in overhead transmission lines,a lightweight detection method MDD-YOLOv3 was proposed.First,the ordinary convolution in the residual unit of the YOLOv3 backbone network is replaced with a depth-wise separable convolution,and the backbone network D-Darknet53 is designed,which greatly improves the detection speed of the network when the network detection accuracy decreases slightly.In the feature mining module,the Dense-SPP module is designed.Dense-SPP and its convolutional feature extraction layer in series can fully mine the global and local features of self-explosion defects,and improve the network’s feature expression ability for self-explosive defects.Finally,a four-dimensional prediction layer is constructed,which can fully extract the location,texture and semantic information of self-explosion defects,and improve the small target detection performance of the network.Simulation experiments show that the detection accuracy of MDD-YOLOv3 for insulators reaches 96.1%,and the detection speed reaches 36 frames per second.Compared with YOLOv3,the detection accuracy and speed are increased by 4.0%and 28.6%,respectively.The research results demonstrate that the proposed method can be adopted to quickly and accurately identify and locate insulator defects in complex backgrounds.