Tiago Sousa

Conducted research on anomaly detection in photovoltaic (PV) modules using multimodal data fusion from UAV aerial inspections. Curated and manually labeled a large dataset with over 120,000 objects for PV panel defect detection using CVAT. Developed early fusion frameworks for combining visual and thermal imagery. Implemented and validated multiple state-of-the-art object detection AI models including RT-DETR and YOLO using PyTorch, Ultralytics, Albumentations, and OpenCV. Achieved a mean average precision (mAP@50) of 94% detecting six types of defects and panels themselves, contributing to advancing PV module maintenance through AI.

Conducted research on anomaly detection for photovoltaic (PV) modules using multi-modal data fusion in aerial inspections. Curated and manually labeled a UAV-based PV inspection dataset with over 120,000 objects using CVAT. Developed an early fusion framework combining visual and thermal images and implemented a framework for training and evaluating object detection AI models including RT-DETR and YOLO using PyTorch, Ultralytics, Albumentations, and OpenCV. Achieved an mAP@50 of 94% for defect detection across six defect categories and PV panels.