Tower Bolt Localization and Missing Detection based on Coordinate Attention Mask and Template-driven Approach

The absence of bolts in transmission towers poses a threat to the safe operation of transmission lines, necessitating regular inspections. Detecting missing bolts is challenging due to the complex background of bolt holes and limited samples. This paper proposes a coordinate attention mask and template-driven method for locating and detecting missing bolts in transmission towers. It can automatically infer missing bolts based on normal bolts on the corresponding components with high accuracy, without relying on samples of missing bolts. Firstly, coordinate attention is introduced into the Mask R-CNN convolutional neural network to identify the tight contour mask of bolts and reduce background interference. Secondly, an adaptive shape feature matching algorithm is proposed to adaptively locate the entire component where the bolt is located around the detection result, and normalize the high-dimensional multi-order moments of the component, bolt, and hole features. The Wasserstein distance is used to compare their distributions with the standard template of the component and its bolt distribution, enabling the localization of missing bolts and completing the detection of their absence. Extensive validation on unmanned aerial vehicle images of a segment of transmission lines in Zengcheng District, Guangzhou City, demonstrates that the proposed algorithm achieves a localization accuracy of $\mathbf{9 5. 7 \%}$ for tower bolts and a detection accuracy of $94.3 \%$ for missing bolts, indicating its practical value.