Enhancing Water Pollutant Detection with few Training Samples using Feature Image and Acoustic Signals

Abstract

Water pollution is a major global environmental issue with severe consequences, highlighting the importance of accurate pollutant detection. While traditional chemical analysis methods are commonly employed to detect water quality, which involves collecting water samples for analysis to detect water quality, they are time-consuming and resource-intensive. In recent years, computer-based detection methods have emerged, but those typically rely on the experience of researchers to manual feature extraction and can be limited in their ability to generalize. This paper proposes a novel approach to water pollutant detection using acoustic signals and Convolutional Neural Network (CNN) technology. By leveraging the microphone and speaker embedded in smartphones to collect and process acoustic data, this method extracts time-frequency domain features and converts them into images for analysis by the CNN. The experimental results demonstrate high accuracy about 98% even in scenarios with limited data, indicating that this method could be an effective and efficient tool for detecting water pollutants.