An infrared image target detection algorithm for porcelain post insulators

Abstract

Detecting abnormal heat in porcelain post insulators using infrared thermal imaging （ITI） stands as a principal approach for diagnosing fault in post insulators within substations. A lightweight target detection model for infrared images of porcelain post insulators is proposed based on computer vision. First， a dilated convolutional kernel is added to the depthwise separable convolution to effectively increase the receptive field of the output unit and reduce the number of parameters. Then， the obtained D-Mobilenet network structure is used to replace the backbone network ELANCSP in YOLOv7， and the SJS （shear， jitter， scale） method is used to expand the number of samples， and algorithms including transfer learning， Mosaic data augmentation， and cosine annealing are introduced to improve the model's generalization ability. Finally， the performance of the model is compared with YOLOv4， YOLOv5， YOLOv7， and the G-Adaboost target detection algorithm. Experimental findings demonstrate that this model boasts a superior combination of lightweight design， robustness， generalization capacity， accuracy， and speed.

Keywords

• porcelain post insulator
• infrared image
• target detection
• lightweight network
• yolov7