Analysis of artificial intelligence methods for rail transport traffic noise detection

Abstract

Nowadays, many cities all over the world suffer from noise pollution. Noise is an invisible danger that can cause health problems for both people and wildlife. Therefore, it is essential to estimate the environmental noise level and implement corrective measures. There are a number of noise identification techniques, and the choice of the most appropriate technique depends upon the information required and its application. Analyzing audio data requires three key aspects to be considered such as time period, amplitude, and frequency. Based on the above parameters, the source of noise can be identified. This research paper suggests the utilization of artificial intelligence and machine learning algorithms for the traffic noise detection process. Computational methods are the fastest and most innovative way to analyze raw data sets and predict results. Identifying patterns in these methods requires a large amount of data and computing power. Machine learning models can be trained using three types of data: experimental sound libraries, audio datasets purchased from data providers, and data collected by domain experts. In the scope of the study, an experimental dataset was used to train a model that predicts the correct outcomes based on the inputs, using supervised learning. Developing an accurate model requires high-quality data input. However, incorrect data collection can cause noise in feature sets, as can human error or instrument error. Traffic sound events in the real environment do not usually occur in isolation but tend to have a significant overlap with other sound events. A part of this paper is dedicated to the problems that may arise during traffic noise detection, like incorrect data processing and data collection. It also discusses the ways to improve the quality of the input data. The study also states that the field of transport noise detection would greatly benefit from the development of a centralized railway database based on constructive railroad data, and from a centralized database with railway-specific datasets. Based on preliminary results of traffic noise analysis, modernization of the tram lines was proposed to reduce the environmental noise.