Akademik Veri Yönetim Sistemi
2nd International Conference on Engineering, Natural Sciences, and Technological (ICENSTED 2025), Bayburt, Türkiye, 19 - 23 Haziran 2025, ss.780-787, (Tam Metin Bildiri)
Photovoltaic
(PV) panels are essential for sustainable energy generation, but their performance
is often compromised by anomalies such as hotspots, microcracks, and surface
contamination. This study introduces an automated fault detection framework
that uses the YOLOv8m deep learning model to localize anomalies in real time in
infrared (IR) thermographic images of PV panels. The dataset, which originally
comprised 2,000 manually labeled IR images, was expanded to 3,600 images using
augmentation techniques, including rotation, flipping, scaling, and color
jitter. After training the model, the proposed system achieved a mean average
precision (mAP@0.5) of 0.7628, a precision of 0.7862, and a recall of 0.6788 on
the test set. These results demonstrate the system’s strong generalization
capability. Visual inspection of the qualitative results confirmed the reliable
detection of prominent and subtle hotspots under diverse environmental and
imaging conditions. The model also performed consistently across varying panel
orientations and color mapping schemes, validating its robustness. The proposed
framework significantly reduces reliance on manual inspection and facilitates
early intervention through fast, accurate anomaly detection. This research
highlights the potential of advanced computer vision systems in supporting
scalable, autonomous PV monitoring and contributes to the broader effort toward
efficient, intelligent renewable energy infrastructure management.