Fully automatic detection and classification of phytoplankton specimens in digital microscopy images

Area of Science:

• Environmental Science
• Microbiology
• Image Analysis

Background:

• Toxin-producing phytoplankton blooms degrade water quality, posing challenges for detection and neutralization.
• Manual phytoplankton analysis is labor-intensive, requires expert knowledge, and suffers from reliability issues.
• Automating phytoplankton analysis is crucial for efficient water quality monitoring and public health protection.

Purpose of the Study:

• To develop a fully automatic methodology for analyzing phytoplankton in digital microscopy images.
• To enable accurate detection, segmentation, and classification of phytoplankton specimens without expert intervention.
• To differentiate phytoplankton from other aquatic particles and identify specific toxic species.

Main Methods:

• A novel automated system for analyzing phytoplankton in digital microscopy images.
• Utilizes a simplified systematic protocol for multi-specimen image acquisition.
• Employs machine learning for specimen detection, segmentation, differentiation from other particles, and classification into target species.

Main Results:

• The system achieved a 0.4% false negative rate (FNR) in detection.
• Phytoplankton detection (differentiating from zooplankton, minerals, etc.) showed 84.07% precision at 90% recall.
• Target species classification accuracy was 87.50%, with high recall for specific toxic species like W. naegeliana (81.82%) and D. sociale (85.71%).

Conclusions:

• The automated methodology provides accurate results in specimen identification, phytoplankton differentiation, and target species classification.
• This fully automatic system offers a robust and consistent tool for water quality analysis.
• The system aids specialists in assessing water source quality and potability, reducing workload and improving reliability.