With the popularization and development of quality education, more and more people realize the importance of music education to the cultivation of students.In order to promote the intelligent development of music education, this paper conducts research on the piano transcription task.A variety of audio signal processing methods are used for feature extraction, and the ResNet18 network is utilized as an implementation method for piano transcription, with improvements such as incorporating the attention mechanism, improving the activation function and using Dropout.Experiments show that the improved design of the ResNet18 network can play a role in the phenomenon of overfitting, and the Loss values of the training set and validation set of the improved ResNet18 network model are 0.025 and 0.11, respectively, which are smaller than those of the ResNet18 network model.The model in this paper performs optimally in the test set comparison experiments, with higher test metric values than other methods on note start + end point, note start + end point + volume, with 1.91%~5.38% and 1.91%~5.38% improvement, respectively.The proposed improved ResNet18 network model has superior piano transcription performance, which can accurately transcribe the music played by the performer to assist teaching work.
