Electric substation inspection: YOLOv5 in hotspot detection through thermal imaging

Main Article Content

Daniel A. Pérez-Aguilar https://orcid.org/0000-0003-4514-2873
Jair. M Pérez-Aguilar https://orcid.org/0009-0009-0875-488X
Andy P. Pérez-Aguilar https://orcid.org/0009-0008-0390-1038
Redy H. Risco-Ramos https://orcid.org/0000-0002-3491-6020
Manuel E. Malpica-Rodriguez https://orcid.org/0000-0002-5534-5116


Substations are key facilities within an electrical system, untimely failures tend to cause low quality and negative effects on the electrical supply. An early indicator of potential electrical equipment failure is the appearance of hot spots; therefore, its detection and subsequent programmed correction avoids incurring in major failures and unnecessary operation stops. In this research, 64 experiments of the YOLOv5 algorithm were carried out, with the purpose of proposing an automated computer vision mechanism for the detection of hot spots in thermal images of electrical substations. The best results show a mAP value of 81.99%, which were obtained with the YOLOv5m algorithm and the transfer learning application. These results leave a basis to deepen and improve the performance of the algorithm by varying other hyperparameters to those considered in this study.