Time: 14:00 - 15:00
Organized by NORA
The motivation for the work is an industrial project where the aim is to monitor power lines from UAV images using specially developed deep convolutional neural networks. The talk describes a proposed neural network called the LS-Net for segmenting out line structures corresponding to power lines. Furthermore, inspired by the power line monitoring problem, the talk describes new basic research within so-called few-shot learning, proposing a new similarity measure to regularize such systems for better performance.