With the successful defence of his doctoral thesis titled “Automated Pipeline Inspection Using Deep Learning-Based Computer Vision“, doctoral candidate Boris Gašparović has officially completed his Doctoral Study Programme.
His research focuses on the development of advanced deep-learning-based computer vision methods for the automatic inspection of sewer, drainage, industrial, and underwater pipelines. As part of his work, he developed standardized datasets with defect annotation procedures based on the EN 13508-2 standard and evaluated multiple generations of YOLO architectures. By incorporating dataset balancing and augmentation techniques, his approach specifically targets the detection of rare and visually challenging defect classes. Additionally, he developed a cascaded YOLO framework in which a lightweight detector performs preliminary defect screening, while a more complex detector is selectively applied to potentially defective images. He also introduced a dual spatiotemporal framework that combines a YOLOv7V-based temporal aggregation stream with the PTSEFormer transformer-based temporal reasoning stream, enabling the integration of spatial and temporal information extracted from inspection videos. The results demonstrate that these proposed methods improve defect detection performance – particularly for underrepresented defect categories – while the cascaded approach significantly reduces computational overhead without sacrificing accuracy compared to conventional single-stage detectors. Overall, the findings confirm the effectiveness of these frameworks for industrial and municipal inspection environments.
The doctoral thesis was supervised by Prof. Jonatan Lerga, and co-supervised by Prof. Goran Mauša (both from the Faculty of Engineering, University of Rijeka).
The Doctoral Thesis Defence Committee consisted of: Prof. Ivo Ipšić (Chair) from the Faculty of Engineering, University of Rijeka, Prof. Marina Ivašić-Kos (Member) from the Faculty of Informatics and Digital Technologies, University of Rijeka, and Assist. Prof. Ivan Markić (Member) from the Faculty of Mechanical Engineering, Computing and Electrical Engineering, University of Mostar.
Congratulations to our colleague!
