Automated identification of toxigenic cyanobacterial genera for water quality control purposes

Authors

Iman Kianian, MohammadSadeq Mottaqi, Fatemeh Mohammadipanah, Hedieh Sajedi

Abstract

Cyanobacteria are the dominating microorganisms in euphotic aquatic environments. The excessive abundance of toxin-producing cyanobacteria cells in drinking water puts public health at substantial risk. Water quality assessments conducted by specialists are time-consuming and also prone to error. In this study, automated identification of toxigenic cyanobacterial genera was conducted with some state-of-art fine-tuned Convolutional Neural Network (CNN) based models, such as MobileNet, MobileNetV2, VGG16, etc. The most successful model for feature extraction from cyanobacterial images was MobileNet. Different methods for classification of the features extracted from CNN models were tested such as Fully Connected Neural Network (FCNN), Support Vector Machine (SVM), Naïve Bayes (NB), eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Multi Layer Perceptron (MLP), and K-Nearest Neighbors (KNN). The experimental results indicate that a FCNN after the convolution layers of fine-tuned MobileNet, could predict the toxigenic genus with weighted accuracy and f1-score of 94.79% and 94.91%, respectively and the highest macro accuracy and f1-score were 90.17% and 87.64% that were gotten with using MobileNetV2 for feature extraction. This approach can considerably boost the effectiveness and accuracy of the action of identifying toxigenic cyanobacteria. Nevertheless, more advanced artificial intelligence (AI)-based systems need to be developed to detect and further be able to predict the concentration of other toxigenic bacteria genera in water samples.