Automatic detection of harmful cyanobacterial genera using deep CNN models and artemisinin optimization

Area of Science:

• Environmental Science
• Microbiology
• Computer Science

Background:

• Global warming exacerbates cyanobacteria blooms, threatening water quality and ecosystem health.
• Current methods for detecting toxic cyanobacteria are slow, labor-intensive, and subjective.
• Harmful cyanobacterial blooms (Cyano-HABs) pose risks to human, animal, and plant life due to toxins and ecological disruption.

Purpose of the Study:

• To develop a novel, automated method for the rapid and precise detection of harmful cyanobacteria genera.
• To address the limitations of traditional cyanobacteria assessment methods.
• To enhance the efficiency and accuracy of water quality monitoring systems.

Main Methods:

• Utilized the TCB-DS dataset for training and validation.
• Employed deep Convolutional Neural Network (CNN) models (ShuffleNet, ResNet-50) for feature extraction.
• Applied feature fusion and the Artemisinin Optimization (AO) algorithm for feature selection.
• Classified cyanobacteria species using multiple CNN models including GoogleNet, MobileNetV2, EfficientNetb0, DarkNet53, ShuffleNet, and ResNet101.

Main Results:

• Achieved a mean accuracy of 97.471% and a maximum accuracy of 97.683% on the TCB-DS dataset.
• The proposed method demonstrated superior performance compared to other models evaluated.
• The Artemisinin Optimization algorithm effectively reduced feature redundancy and improved model efficiency.

Conclusions:

• The novel automated detection method significantly enhances water quality monitoring capabilities.
• This approach offers a fast, precise, and objective alternative to traditional cyanobacteria assessment.
• The findings contribute to mitigating the risks associated with increasing harmful cyanobacterial blooms globally.