Learning Diatoms Classification from a Dry Test Slide by Holographic Microscopy.

Area of Science:

• Marine biology and environmental science
• Computational intelligence and machine learning

Background:

• Diatoms are key phytoplankton and water quality biomarkers, but their classification is challenging.
• Current diatom taxonomy relies on expert experience, which is time-consuming and subjective.
• Deep learning excels at image classification but requires extensive training data, difficult to obtain for diverse microalgae.

Purpose of the Study:

• To develop an accurate and efficient method for diatom classification using deep learning.
• To leverage commercial diatom test slides and data augmentation to overcome data limitations.
• To validate the deep learning model using holographic imaging of diatoms in natural conditions.

Main Methods:

• Utilized commercial diatom test slides with 50 fixed species for dataset creation.
• Employed data augmentation techniques to expand the training dataset from single images.
• Developed a deep Convolutional Neural Network (CNN) ensemble for classification.
• Incorporated holographic imaging for quantitative phase-contrast mapping and 3D imaging capabilities.
• Validated the model with holographic recordings of live diatoms in water samples.

Main Results:

• Successfully trained accurate deep Convolutional Neural Networks (CNNs) for diatom classification.
• Demonstrated the efficacy of using augmented datasets derived from commercial test slides.
• Achieved accurate classification of 50 diatom species.
• Validated the model's performance on live diatoms in their natural aquatic environment.

Conclusions:

• Commercial diatom test slides are highly effective for training accurate deep learning models.
• This approach significantly reduces the time and expertise required for diatom identification.
• The developed method offers a streamlined and reliable solution for environmental monitoring and water quality assessment.