Ornaments and Barlines Recognition of Numbered Musical Notation Using YOLOv5

Abstract

Ornament and barline of numbered musical notation recognition is an important branch in optical music recognition (OMR), traditional template matching methods are computationally intensive and highly dependent on templates, resulting in weak robustness and poor generalization. We present a case study of how You Only Look Once (YOLO) v5 can be implemented to detect ornament and barline notations in numbered musical notation in order to improve the rate and accuracy of ornament and barline detection in optical music recognition. There are no sketch sheet music sets in the public domain, so we created our own training set. We applied transfer learning to YOLOv5 to investigate the appropriate features for the recognition of ornament and barline. Our results show that the trained algorithm can detect ornament and barline notations in numbered musical notation with high confidence.