Based on a systematic review of convolutional neural networks (CNN), this study explores the efficacy of small imaging sensors in monitoring the real-time presence of cyanotoxins and hazardous contaminants in urban ecosystems. To develop a machine learning-based CNN, this study first investigated the relationships between the prevalence of hazardous algal blooms (HABs) and faecal indicator bacteria (FIB) in waterways and aquifers of certain semi-arid zones of Sri Lanka, Sweden and New York (United States). By incorporating a popularly known AbspectroscoPY framework to effectively process the spectrophotometric data of the obtained samples, the formulation subsequently reveals strong positive correlations between FIB coliforms and nutrient loads (particularly nitrate and phosphate). A corroborative association with the incidence of chronic kidney disease of uncertain aetiology (CKDu) among the residents of the studied regions further affirms the reliability of the methodology. These findings underline the need for policymakers to consider the geographical and land-use traits of urban habitats in strategies aimed at reducing water-borne health hazards. HIGHLIGHTS This study examines the link between hazardous algal blooms (HABs) and faecal indicator bacteria (FIB) in the semi-arid habitats of Sri Lanka, Sweden, and New York, USA. Quantitative Phase Imaging based on a convolutional neural network (CNN) model helps monitor cyanobacterial incursions in peri-urban agrarian ecosystems. AbspectroscoPY-enabled spectrophotometric data analysis reveals strong positive correlations between the prevalence of FIB coliforms and nutrient loads, particularly those of nitrates and phosphates. Reliability of proposed machine learning-based CNNs is validated by the corroborative incidences of chronic kidney diseases among residents of the studied regions. GRAPHICAL ABSTRACT
Read full abstract