MYCOLab Ai Logo - gold and army green

Hybrid Deep Learning Framework for High-Accuracy Classification of Morphologically Similar Puffball Species Using CNN and Transformer Architectures

Summary

Scientists developed an artificial intelligence system that can automatically identify eight different types of puffball mushrooms from photographs with 95% accuracy. The study compared five different AI models and found that a modern convolutional neural network called ConvNeXt-Base was the best at telling apart puffball species that look very similar to each other. This technology could help amateur mushroom enthusiasts, researchers, and nature conservationists accurately identify these fungi without needing a microscope or laboratory tests.

Background

Puffballs are a group of macrofungi belonging to Basidiomycota that pose significant taxonomic challenges due to convergent morphological features including spherical basidiocarps and similar peridial structures. Traditional identification methods are labor-intensive and prone to misclassification, particularly among morphologically similar species. Recent advances in deep learning offer potential solutions for automated and accurate fungal species identification.

Objective

This study proposes a comprehensive deep learning-based classification framework for eight ecologically and taxonomically important puffball species. The objective is to comparatively evaluate five state-of-the-art deep learning architectures (ConvNeXt-Base, Swin Transformer, Vision Transformer, MaxViT, and EfficientNet-B3) to determine their effectiveness in species-level discrimination among visually similar fungal taxa.

Results

ConvNeXt-Base achieved the highest performance with 95.41% accuracy, 0.96 precision, 0.95 recall, and 0.95 F1-score. Swin Transformer followed with 92.08% accuracy. Vision Transformer and MaxViT showed moderate performance (83.75-84.16% accuracy), while EfficientNet-B3 lagged with 82.08% accuracy. ConvNeXt-Base excelled at distinguishing morphologically similar species, achieving near-perfect AUC values (0.99-1.00) across all eight puffball species.

Conclusion

Deep learning models, particularly ConvNeXt-Base, demonstrate powerful capabilities for accurate species-level classification of morphologically similar puffball fungi. The framework successfully addresses fine-grained taxonomic challenges and shows promise for automated mushroom identification systems. Future applications could expand to mobile platforms and citizen science initiatives to democratize fungal biodiversity monitoring and support conservation efforts.
  • Published in:Biology (Basel),
  • Study Type:Comparative Evaluation Study,
  • Source: PMID: 40723375, DOI: 10.3390/biology14070816
Scroll to Top