Improved insulator location and defect detection method based on GhostNet and YOLOv5s networks.

Outdoor, real-time, and accurate detection of insulator defect locations can effectively avoid the occurrence of power grid security accidents. This paper proposes an improved GhostNet-YOLOv5s algorithm based on GhostNet and YOLOv5 models. First, the backbone feature extraction network of YOLOv5 was reconstructed with the lightweight GhostNet module to reduce the number of parameters and floating point operations of the model, so as to achieve the purpose of being lightweight. Then, a 160 × 160 feature layer was added to the YOLOv5 network to extract more feature information of small targets and fuzzy targets. In addition, the introduction of lightweight GSConv convolution in the neck network further reduced the computing cost of the entire network. Finally, Focal-EIoU was introduced to optimize the CIoU bounding box regression loss function in the original algorithm to improve the convergence speed and target location accuracy of the model. The experimental results show that the parameter number, computation amount, and model size of the GhostNet-YOLOv5s model are reduced by 40%, 25%, and 36%, respectively, compared with the unimproved YOLOv5s model. The proposed method not only ensures the precision of insulator defect detection, but also greatly decreases the complexity of the model. Therefore, the GhostNet-YOLOv5s algorithm can meet the requirements of real-time detection in complex outdoor environments. [ABSTRACT FROM AUTHOR]