EU PVSEC Programme Online
EU PVSEC 2021, 6 - 10 September 2021
Presentation: 5CV.2.33 Deep Learning Based Object Detection Algorithm for PV Module Defects
Type: Visual
Date: Wednesday, 8th September 2021
13:30 - 15:00
Author(s): S. Xu, M. Ziyao
Presenter / Speaker: M. Ziyao, Nankai University, Tianjin, China
Event: Conference Conference
Session: 5CV.2 Operation, Performance and Maintenance of PV Systems
Topic: 5. 3 Operation, Performance and Maintenance of PV Systems
Summary / Abstract: To address the low efficiency and error-prone using manual inspection of PV module defects. This paper uses YOLO V4, a object detection algorithm based deep learning, to detect defective cells on the EL image of photovoltaic modules. Experimental results show that the detection method is very fast (an EL image takes only 54.71ms), while the AP (Average precision) can reach 81.66%. In order to make better use of the collected data, this work has carried out data enhancement, such as adjusting saturation, exposure, hue of image and performing cutmix and mosaic operations. Experiments show that the data enhancement operation can improve the model AP by 83.42% without increasing any calculation quantity. This method is of great significance for PV module inspection in production lines and photovoltaic power plants.