Deep Learning-based Method for PV Panels Segmentation and Defects Detection with Infrared Images

The health condition evaluation of photovoltaic plants is considered a significant challenge for years. This paper proposed a framework for photovoltaic panels segmentation and defects detection in module-level using infrared Images through addressing three technical challenges: (1) providing some high-quality infrared images captured by Unmanned Aerial Vehicles (UAV) during the inspection process; (2) an adaptive U
²
-Net network is designed to locate the PV panels in the aerial images; (3) YOLOv4 is introduced to detect the PV defects in module-level. With the support of multi-scene data, the framework is evaluated by numerical experiments using the existing approaches as comparison benchmarks. The extensive results confirmed the efficiency and accuracy of the framework proposed in this paper.