Direct detection and classification of E. coli and fecal streptococci using an optical water droplet method and convolutional neural networks

Abstract

Detecting harmful bacteria in drinking water is a significant concern for public health. Indicator bacteria like E. coli and fecal streptococci serve as markers for fecal contamination of water [Environmental and Pollution Science (2019), pp. 191]. Several detection methods of these markers require costly equipment, and specialized laboratories and technicians, and are time-consuming, resulting in labor-intensive processes [J. Phys. Conf. Ser.995, 012065 (2018)1742-658810.1088/1742-6596/995/1/012065]. This paper proposes an optimal method that combines an optical water droplet method and convolutional neural networks (CNNs) to provide an accurate and cost- and time-effective approach to directly detect and classify E. coli and fecal streptococci in water. The system captures images of indicator bacteria and then classifies them. We obtain a classification accuracy of up to 0.89 and a loss of 0.13. This work constitutes a step toward an integrated, real-time, and automatic optical-based detection system for water-borne pathogenic agents.