Detection of Discharging Insulators via Acoustic and Visible Image Fusion

Insulators are crucial components in power systems, and timely detection of discharge defects is essential for ensuring reliable performance. The noise source localization and image processing algorithms help to identify insulators experiencing partial discharges, but each approach has its own strengths and limitations. To tackle it, this article proposes a method for detecting insulators with partial discharge based on acoustic and visible image fusion. First, a dataset composed of acoustic maps is constructed using simulation and the generalized cross-correlation phase transform (GCC-PHAT). Next, pyramid structured acoustic and visual feature extraction networks are used to capture multiscale features from both modalities. Subsequently, a detection head, composed of multiple acoustic attention modules and anchor box initialization modules, is employed to fuse visual and acoustic features. The specialized design of the detection head allows it to guide attention toward areas generating noise. The experimental results show that the proposed model achieves an AP50 of 93.4% and 92.2% on low-quality datasets, surpassing other classic models. In addition, the modules in the proposed model can be easily extended to other domains where more attention is required for noise-generating areas.