Traffic-noise Annoyance Assessment Method Combining LSTM and Transfer Learning

This study is dedicated to developing a deep learning-based traffic noise signal evaluation method, with the goal of improving the accuracy and efficiency of noise evaluation and providing solid scientific support for urban planning and noise management. With the acceleration of urbanization, the interference of traffic noise on residents' daily lives is becoming increasingly significant, and the evaluation of traffic noise has gradually received widespread attention from researchers. This paper proposes a traffic noise annoyance assessment model that combines Long Short Term Memory Network (LSTM) and transfer learning. This method elaborates in detail on the working principle, structural design, selection of loss function, configuration of optimizer during training, and adjustment strategy of learning rate of the model. The experimental results show that compared with other methods, the evaluation model in this paper performs well in evaluation indicators such as mean absolute error, Pearson correlation coefficient, and Spearman correlation coefficient. In addition, the evaluation performance of the model in different annoyance degree intervals also reaches a high level.