Multiple marine algae identification based on three-dimensional fluorescence spectroscopy and multi-label...

Area of Science:

• Marine biology
• Spectroscopy
• Machine learning

Background:

• Accurate identification of marine algae is crucial for monitoring seawater quality.
• Fluorescence techniques are effective but challenged by similar pigments in coexisting algae.
• Existing methods struggle with complex algal mixtures.

Purpose of the Study:

• To develop and validate a multi-label classification model for precise identification of single and mixed algal samples.
• To improve the efficacy of fluorescence-based methods for algal identification in complex marine environments.
• To assess the performance of a novel EEM-CNN model combined with 3D fluorescence spectroscopy.

Main Methods:

• Developed a multi-label classification model integrating a specific Excitation-Emission Matrix Convolutional Neural Network (EEM-CNN) with 3D fluorescence spectroscopy.
• Utilized rectangular convolutional kernels and double convolutional layers for enhanced spectral feature extraction.
• Trained and validated the model on a dataset of 3D fluorescence spectra from eight algae species, including augmented and test samples.

Main Results:

• Achieved a classification accuracy of 0.883 and an F1 score of 0.925 on 4448 training and 60 test samples.
• The EEM-CNN model demonstrated superior recognition accuracy for both single and mixed algal samples compared to ML-kNN and N-PLS-DA.
• The model showed robust performance across varying sample concentrations and growth stages, even with spectral similarity.

Conclusions:

• The developed EEM-CNN model combined with 3D fluorescence spectroscopy offers a promising tool for precise marine algae identification.
• This approach enhances the capability of fluorescence-based techniques for complex seawater quality monitoring.
• The model's robustness highlights its potential for real-world applications in marine ecological studies.