Please use this identifier to cite or link to this item: http://hdl.handle.net/2080/4671
Full metadata record
DC FieldValueLanguage
dc.contributor.authorPanigrahy, Satyajit-
dc.contributor.authorSahoo, Raseswar-
dc.contributor.authorKarmakar, Subrata-
dc.date.accessioned2024-09-06T11:21:22Z-
dc.date.available2024-09-06T11:21:22Z-
dc.date.issued2024-08-
dc.identifier.citationIEEE International Conference on High Voltage Engineering and Application (ICHVE), Berlin, Germany, 18-22 August 2024en_US
dc.identifier.urihttp://hdl.handle.net/2080/4671-
dc.descriptionCopyright belongs to proceeding publisheren_US
dc.description.abstractThis study proposes a single-stage deep learning model capable of condition classification, defect detection, and segmentation for high-voltage insulator inspection. Leveraging the Insulator Defect Image Dataset (IDID) from the Electric Power Research Institute (EPRI), which includes three conditions (good, broken, and flashed surface), the model undergoes image pre-processing before training. Utilizing the auto-annotation feature of the Segment Anything (SAM) model, a segmentation dataset is generated using a pre-trained YOLOv8 detection model. The experimental results demonstrate exceptional performance, with the insulator condition classification model achieving 93.49% accuracy and the detection and segmentation models achieving mAP@50 of 92.9% and 82.6%, respectively. This remarkable performance enables proactive maintenance, minimizes downtime, and enhances power systems’ overall security and reliabilityen_US
dc.subjectOutdoor Insulatoren_US
dc.subjectCondition Classificationen_US
dc.subjectDefect Detectionen_US
dc.subjectSegmentationen_US
dc.subjectSegment Anything (SAM)en_US
dc.subjectYOLOv8en_US
dc.titleInsulator Condition Classification, Defect Detection, and Segmentation using Yolov8 Deep-Learning Modelen_US
dc.typeArticleen_US
Appears in Collections:Conference Papers

Files in This Item:
File Description SizeFormat 
2024_ICHVE_Spanigrahy_Insulator.pdf2.92 MBAdobe PDFView/Open    Request a copy


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.