Application of Deep Learning Based Detector YOLOv5 for Soiling Recognition in Photovoltaic Modules

Abstract

Objects detection algorithms based on Convolutional Neural Networks (CNN) are becoming dominant in many application fields due to their higher accuracy advantage over traditional algorithms. YOLO framework is one of the most popular algorithms used for object detection and has shown a comparatively similar performance to the R-CNN algorithms. This article presents real-time soiling and dust accumulation detection on photovoltaic (PV) panels based on the YOLOv5 framework. The proposed method aims to address several challenges related to the evaluation and detection of soiling status in photovoltaic systems and also offers a solution to the problem of the lack of effective evaluation methods and automatic detection technologies. It can offer several advantages over traditional methods to significantly improve the accuracy and efficiency of real-time soiling status inspection of photovoltaic panels by introducing effective evaluation techniques and utilizing automatic detection technologies. A PV system included a standalone PV module of four panels of Poly-crystalline silicon was used to validate the corresponding approach. The proposed method performed an accuracy of more than 80%. This approach can be used to build a high-performance real-time embedded system for dust and soiling recognition in photovoltaic panels based on CNNs and computer vision techniques.