Kuva: Christine Grafström
In his doctoral dissertation, M.Sc. (Computer Science) Ali Hassan investigated how deep-learning–based image analysis can be made more parameter-efficient without sacrificing accuracy. Modern image-recognition systems often rely on large and computationally expensive neural networks, which limits their use in real-world and resource-constrained environment. This research focuses on redesigning convolutional neural network architectures to reduce the number of parameters and computational cost while maintaining model accuracy. The proposed methods were evaluated across multiple computer-vision applications, including image classification, fire segmentation, and disparity estimation from light-field images. The results show that the proposed lightweight networks require up to 76% fewer parameters while achieving accuracy comparable to state-of-the-art models, enabling advanced deep learning more accessible for image-analysis solutions.
The doctoral dissertation of M.Sc. (Computer Science) Ali Hassan in the field of Plenoptic Imaging titled Parameter-Efficient Convolutional Neural Networks for Computer Vision Applications will be publicly examined at the Department of Computer and Electrical Engineering at Mid Sweden University, Sweden, on 27 February 2026.
The Opponent will be Professor Jenny Benois Pineau from the Université de Bordeaux in France. The Custos will be Professor Mårten Sjöström from the Department of Computer and Electrical Engineering, Mid Sweden University, Sweden.