Comparative assessment of artificial intelligence (AI)-based algorithms for detection of harmful bloom-forming algae: an eco-environmental approach toward sustainability

Abstract

Abstract Organic effluent enrichment in water may selectively promote algal growth, resulting in water pollution and posing a threat to the aquatic ecosystem. Recent harmful algal blooms (HABs) incidents have highlighted information gaps that still exist, as well as the heightened need for early detection technology developments. Although previous research has demonstrated the importance of deep learning in the identification of algal genera, it is still a challenge to identify or to develop the best-suited convolution neural network (CNN) model for effective monitoring of bloom-forming algae. In the present study, efficiency of deep learning models (MobileNet V-2, Visual Geometry Group-16 (VGG-16), AlexNet, and ResNeXt-50) have been evaluated for the classification of 15 bloom-forming algae. To obtain a high level of accuracy, different convolution layers with adaptive moment estimation (Adam), root-mean-square propagation (RMSprop) as optimizers with softmax and rectified linear unit (ReLU) as activation factors have been used. The classification accuracies of 40, 96, 98, and 99% have been achieved for MobileNet V-2, VGG-16, AlexNet, and ResNeXt-50 model, respectively. We believe that the ResNeXt-50 has the potential to identify algae in a variety of situations with high accuracy and in real time, regardless of the underlying hardware. Such studies pave the path for future AI-based cleaner technologies associated with phycological studies for a sustainable future.

Keywords

• Harmful algal blooms (HABs)
• Convolution neural network (CNN)
• Cleaner technologies
• Convolution layers
• Deep learning models