Image-based Efficient Music Note Detection by Advanced Lightweighted YOLO

Abstract

In the realm of music note detection, deep learning models have demonstrated exceptional potential, enabling a wide range of applications, including automated music transcription and real-time music analysis. Despite their effectiveness, the resource-intensive nature of these models often renders them impractical for deployment in resource-limited environments. To tackle this issue, this research focuses on adapting an existing deep learning model, You Only Look Once (YOLO), to explore lightweight alternatives specifically tailored for music note detection. This work assessed the performance of the adapted model using a comprehensive dataset that includes 2,136 meticulously labeled music sheet images. Preliminary results suggest that the streamlined version of the model not only matches the accuracy of its predecessor but also boasts a substantially lower parameter count and reduced Floating Point Operations (FLOPs). These enhancements make the model an ideal candidate for music note detection across a broader spectrum of devices, including those with limited computational capabilities. This advancement opens up new possibilities for real-time music analysis and transcription in various settings, such as mobile devices and low-power embedded systems.