Intelligent Fault Pattern Recognition of Aerial Photovoltaic Module Images Based on Deep Learning Technique

Xiaoxia Li, Qiang Yang, Wenjun Yan, Zhebo Chen

The rise of photovoltaic industry has raised the difficulty of the operation and maintenance. Nowadays, the growing interest in the application of unmanned aerial vehicles (UAV) in civil monitoring and diagnostic applications has been observed. Such UAV-based inspection system can significantly improve the efficiency of system monitoring and fault detections. This paper presents an intelligent UAV-based inspection system for asset assessment and defect classification for large-scale PV systems. The aerial imagery data of PV modules increase the complexity of the detection by traditional pattern recognition, a novel method based on the deep learning and supervision is proposed, which could solve the low quality and distortion flexibly and reliably. A convolutional neural network (CNN) is adopted to address the defects classification. Extracting features by the pre-trained architecture Vgg16, the suggested solution added a full-connected layer and a SVM decision layer to classify the defects. Such pre-trained learning-based algorithm can meet the demand of the small datasets, and carry out a variety of deep features and condition classification in PV system, which can supervise with significantly promoted efficiency in comparison with the conventional methods. The proposed solution is evaluated through numerical experiments and the result confirms its improved performance.

Full Text