Muhammad Haroon

Presents an end-to-end railway track defect detection system combining a supervised YOLOv8x segmentation model and a self-supervised U-Net for rail surface anomaly segmentation; reports high mAP and real-time inference performance and discusses dataset curation and training strategies to handle limited labeled data.

Research on multisized object detection in spaceborne optical imagery, reflecting work in aerial surveillance, neural networks, and satellite imagery.