DETECTION OF WASTEWATER POLLUTION THROUGH NATURAL LANGUAGE GENERATION WITH LOW-COST SENSING PLATFORM

Abstract

In order to safeguard individuals and anticipate potentially harmful situations, it is crucial to detect 
pollutants in various settings, such as air, water, and sewage systems. The majority of these studies use 
traditional Machine Learning techniques to process the collected measurement data. First, a novel 
classification approach to classify contaminants in wastewater based on deep learning and the 
transformation of raw sensor data into natural language metadata; second, a low-cost platform to 
acquire, pre-process, and transmit data for this purpose. Superior efficacy and tolerable efficiency 
distinguish the suggested solution from state-of-the-art methods. Knowing the injection time—the 
precise moment when the pollutant is introduced into the wastewater—is crucial to the suggested 
method, which is its primary drawback. Therefore, a finite state machine tool capable of inferring the 
precise moment of injection is also included into the proposed system. There is an extensive presentation 
and discussion of the complete system. In addition, we provide many versions of the suggested 
processing method to evaluate how the system responds to changes in the amount of samples and the 
accompanying computing load and speed. While the best baseline approach only managed an accuracy 
of 81.0%, our strategy achieved a minimum of 91.4%.