This thesis addresses the problem of automated detection and segmentation of components in an experimental setup consisting of beams and accelerometers, using machine vision. To tackle this, a transfer learning approach was employed by fine-tuning a YOLOv11s-seg model on a custom made, manually annotated dataset. The model’s performance was evaluated using standardized metrics. Results demonstrate high accuracy and reliable segmentation on unseen test images, confirming strong generalization capabilities. The proposed approach proved effective for automating the identification of key experimental components in laboratory environments.
|
