Improving Small Scale Electric Arc Detection in Railway Systems via Pantograph Centered ROI Extraction
8th Global Power, Energy and Communication Conference, GPECOM 2026, Naples, İtalya, 3 - 05 Haziran 2026, ss.911-916, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/gpecom70462.2026.11578838
- Basıldığı Şehir: Naples
- Basıldığı Ülke: İtalya
- Sayfa Sayıları: ss.911-916
- Anahtar Kelimeler: electric arc detection, pantograph-catenary system, region of interest (ROI), small-object detection
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
In electric railways, the pantograph is the only connection between the moving train and the overhead line. When the electric arc occurs at the contact point, this means that there may be a problem with the system, it may lead to damage to the conductor. The aim of this research is to propose an extraction framework of pantograph-centered dynamic region of interest (ROI), which constrain the detection of the area of physical contact. The performance of three different models (YOLOv11n, Faster R-CNN and RetinaNet) using full frame imagery was assessed for their detection capabilities. YOLOv11n was the only model to exceed the baseline performance of the other models once it was developed into an appropriately defined ROI based framework. When comparing to one another, YOLOv11n mean average precision (mAP)@0.5:0.95 improved by 19.4 points (from 38.58% to 57.96%) by using the ROIs to train the model. Therefore, the use of a spatial constraint on the detection region is beneficial in improving the accuracy of detecting small size arcs when monitoring railways.