Instance Segmentation Graph Neural Network for Detecting Faulty Insulators Using the FLIR SkyRanger R70 UAS

This paper explores the use of Unmanned Aerial System (UAS) for detecting faulty insulators using a Graph Neural Network called Instance Segmentation Graph Neural Networks (IS-GNN). Research regarding the problem of faulty insulator detection on power distribution networks remains an open research area, due to the significant role insulators play to maintain quality inspection in transmission of power lines. Most existing methods detect insulators by rectangular bounding box but do not perform segmentation down to the instance pixel-pixel level. In this paper, a novel Instance Segmentation Graph Neural Network (IS-GNN) method for detecting faulty insulators is proposed. Specifically, graph nodes representing instance features are used for detection and segmentation while graph edges representing instance relations used for tracking. Both inter and infra-frame information is effectively propagated and shared via graph updates and all the sub problems (i.e. detection, segmentation and tracking) are jointly optimized in an unified framework. Existing methods either rely only on single-frame information for the detection and segmentation sub problems or handle tracking as a separate post-processing step, which limit their capability to fully leverage and share useful spatial temporal information for all the sub problems. This IS-GNN model is trained on a custom data set and deployed on to a UAS for evaluating the model in real-time.