Research Internship on Line Pylons Asset Management

Project Outcomes

During this internship, I collaborated with the project engineers to design a detection and classification algorithm for insulating ceramics based on the YOLOv8 framework, which improved the detection accuracy of the original FastRCNN-based system by 10.2%. Together, we also developed an uncertainty-based active learning solution for managing electrical pylon assets, achieving performance comparable to full-data training while using only 40% of the labeled dataset, thereby significantly reducing data collection, annotation, and training costs. The project concluded with the joint preparation of a technical report.