A COLLABORATIVE SYSTEM FOR GEOREFERENCED MONITORING OF NOISE POLLUTION DATA CLASSIFIED BY MACHINE LEARNING MODEL

This paper proposes a collaborative system for the georeferenced monitoring of noise pollution in urban environments, utilizing a machine learning model for automatic sound classification. The system adopts a modular architecture composed of three main applications: a Client Application (collection), a Server Application (processing), and a Client Application (visualization), which allows for scalability and ease of maintenance. In the collection Client Application, data capture is performed via a chatbot integrated with the Telegram application, enabling users to submit audio recordings along with their locations. These data are processed by a Server Application, which extracts acoustic features, classifies the sounds using the UrbanSound8K_ECAPA model, and stores the results in a NoSQL database. The visualization Client Application consists of a web page that requests data from the Server Application and presents the captured data interactively, allowing for the exploration of classified sounds and their corresponding geographical locations. A comparative analysis of audio classification models was conducted to support the choice of the most suitable model. Preliminary results demonstrate the viability of the proposed solution, which proves to be a promising, low-cost, accessible, and scalable tool for mapping and analyzing urban noise pollution, with potential applications in public policy and smart city initiatives.