Transfer Learning based Photovoltaic Module Defect Diagnosis using Aerial Images

The automated system inspection with defect detection in large-scale photovoltaic (PV) farms is an urgent problem to be addressed. This paper presents a transfer learning based solution for visible module defects diagnosis in these conditions. The solution mainly includes three parts: a normative, time-updated, sundry dataset; a pre-trained deep learning model; a transfer learning strategy. Image augmentation technology is applied in the dataset. The proposed transfer learning based solution is able to extract the defects' features from local representation to global one and low level to high one with more robustness in comparison with the enhanced CNN based method. The proposed solution is evaluated through extensive experiments and the numerical results clearly demonstrate its effectiveness.