期刊名称:Journal of Systemics, Cybernetics and Informatics
印刷版ISSN:1690-4532
电子版ISSN:1690-4524
出版年度:2018
卷号:16
期号:2
页码:67-71
出版社:International Institute of Informatics and Cybernetics
摘要:The rise of photovoltaic industry has raised the difficulty of theoperation and maintenance. Nowadays, the growing interest inthe application of unmanned aerial vehicles (UAV) in civilmonitoring and diagnostic applications has been observed.Such UAV-based inspection system can significantly improvethe efficiency of system monitoring and fault detections. Thispaper presents an intelligent UAV-based inspection system forasset assessment and defect classification for large-scale PVsystems. The aerial imagery data of PV modules increase thecomplexity of the detection by traditional pattern recognition, anovel method based on the deep learning and supervision isproposed, which could solve the low quality and distortionflexibly and reliably. A convolutional neural network (CNN) isadopted to address the defects classification. Extractingfeatures by the pre-trained architecture Vgg16, the suggestedsolution added a full-connected layer and a SVM decision layerto classify the defects. Such pre-trained learning-basedalgorithm can meet the demand of the small datasets, and carryout a variety of deep features and condition classification in PVsystem, which can supervise with significantly promotedefficiency in comparison with the conventional methods. Theproposed solution is evaluated through numerical experimentsand the result confirms its improved performance.
关键词:UAV; PV inspection; CNN; and Defects;classification.