Environmental Sound Pollution Mapping Using Supervised Deep Learning

An accurate sound pollution map is necessary in urban environments where sound pollution is prevalent. However, obtaining dense sound pollution data across an area is challenging due to sparse measurements and limited availability of real data. This project aims to develop a framework for predicting dense sound pollution maps based on sparse measurements, leveraging techniques from computer vision. A U-Net Regression model is employed to reconstruct dense sound pollution maps from sparse measurements, portraying the model as a technique for generating sound pollution map images. The performance of the U-Net model is compared with the popular mapping technique called Gaussian Process Regression (GPR), demonstrating that the U-Net model achieves superior accuracy. Furthermore, the ability to generate dense sound pollution maps from sparse data collection via sensors or other methods makes our approach economically and environmentally sustainable.