Smartphone-Based Microalgae Monitoring Platform Using Machine Learning

Abstract

There is a growing demand for microalgae monitoring techniques since microalgae are one of the most influential underwater organisms in aquatic environments. Specifically, such a technique should be hand-held, rapid, and easily accessible in the field since current methods (benchtop microscopy, flow cytometry, or satellite imaging) require high equipment costs and well-trained personnel. This study’s main objective was to develop a field-deployable microalgae monitoring platform using only a single smartphone and inexpensive acrylic color films. It aimed to evaluate the morphological states of microalgae including stress, cell concentration, and dominant species. Using a smartphone’s white LED flash and camera, the platform detected fluorescence and reflectance intensities from microalgal samples in various excitation and emission color combinations. Multidimensional intensity data were evaluated from the smartphone images and used to train a support vector machine (SVM) based machine learning model to classify various morphological states. The SVM classification accuracies were 0.84-0.96 in classifying four- to five-tier stress types, cell concentration, and dominant species and 0.99-1.00 in classifying two-tier stress types and cell concentrations. Additional field samples were collected from the local pond and independently tested using the laboratory-collected training set, showing two-tier classification accuracies of 0.90-1.00. This platform enables accessible and on-site microalgae monitoring for nonexperts and can be potentially applied to monitoring harmful algal blooms (HABs).