Key target and defect detection of high-voltage power transmission lines with deep learning

Abstract

Automatic visual detection of key targets and defects for power transmission lines based on power transmission line inspection robots (PTLIR) and unmanned aerial vehicles (UAVs) is an ongoing trend in the smart grid development. The advancement of deep learning has accelerated the intelligence of power grid inspection. In terms of the power transmission line fittings detection application, improving the detection accuracy of small targets and defects using a deep learning detection network is challenging, owing to the complex background lighting characteristics of high-voltage power transmission lines in the wild. To address this problem, we present an inspection method for key targets and defects in high-voltage power transmission lines based on a deep learning object detection network. First, we collected sample images of key targets under different backgrounds, lighting conditions, and postures. Further, data augmentation was performed to solve the problem of imbalance in the number of target categories, and a large standard dataset was constructed. Second, we improved the extraction ability of small object features by optimizing the detection network. The precision and recall rate of the optimized detection network were 93.5% and 96.2%, respectively. Furthermore, small targets and defects in a complex environment could be successfully detected. Additionally, the detection of targets and defects in the inspection videos recorded by the PTLIR and UAVs were realized. Experimental results demonstrated that the proposed method performed well in the detection accuracy of key targets and defects in similar high-voltage power transmission line environments. It can realize remote, automatic inspection of high-voltage power transmission lines in the field.