Explainable Deep Learning Framework for Reliable Species-Level Classification Within the Genera Desmodesmus and...

Area of Science:

• * Algology and Computational Biology: Focuses on the classification of microalgae using advanced computational techniques.
• * Eukaryotic Photosynthesis: Investigates photosynthetic eukaryotes, specifically green microalgae.

Background:

• * Microalgae exhibit significant morphological diversity, complicated by environmental plasticity, making taxonomic resolution challenging.
• * Accurate classification is crucial for ecological monitoring, biomass optimization, and biofuel production.

Purpose of the Study:

• * To develop a transparent, reliable framework for green microalgae (Chlorophyta) classification by merging deep learning and explainable artificial intelligence (XAI).
• * To evaluate the performance and biological relevance of various deep learning models for microalgal species identification.

Main Methods:

• * Analysis of 3624 microscopic images from three Chlorophyta species using twelve convolutional neural networks (e.g., EfficientNet, ResNet152V2).
• * Standardized preprocessing and data augmentation, including contrast enhancement and normalization.
• * Performance assessment using accuracy and F1-score; interpretability evaluation via saliency maps and Grad-CAM.

Main Results:

• * ResNet152V2 achieved the highest performance, outperforming other architectures in macro F1-score.
• * XAI visualizations (saliency maps, Grad-CAM) confirmed models focused on biologically relevant features (cell walls, ornamentation).
• * High test accuracy was achieved even with constrained datasets, demonstrating model reliability.

Conclusions:

• * The integration of deep learning and XAI provides a robust method for automated microalgal taxonomy.
• * This approach supports biodiversity monitoring, ecological assessment, and the optimization of biomass and biodiesel production.
• * The study highlights the utility of interpretable AI in biological classification tasks.