MYCOLab Ai Logo - gold and army green

A Deep Learning and Explainable AI-Based Approach for the Classification of Discomycetes Species

Summary

Researchers developed an artificial intelligence system to automatically identify 14 different types of cup fungi (Discomycetes) from photographs. Using a method called EfficientNet combined with explainable AI, the system achieved 97% accuracy in species identification. The technology could help scientists quickly and accurately catalog fungal biodiversity for conservation efforts and ecological studies without requiring expert mycologists to examine every specimen.

Background

Discomycetes comprise over 9000 diverse fungal species characterized by cup- or disc-shaped apothecia. Modern taxonomy requires phylogenetic data for accurate classification rather than morphological traits alone. This study addresses the need for automated, accurate species identification using advanced computational methods.

Objective

To develop and evaluate deep learning models combined with explainable AI techniques for classifying 14 distinct Discomycetes species from macrofungal images. The study aims to improve traditional taxonomy through automated image analysis while ensuring model transparency and interpretability.

Results

EfficientNet-B0 achieved the highest performance with 97% accuracy, 97% F1-score, and 99% AUC. MobileNetV3-L followed with 96% accuracy and 99% AUC, while ShuffleNet reached 95% accuracy. EfficientNet-B4 showed lower performance at 89% accuracy, demonstrating that model efficiency rather than complexity determines success under dataset constraints.

Conclusion

The study demonstrates that well-optimized lightweight CNN architectures outperform larger models when training data is limited. EfficientNet-B0 and MobileNetV3-L provide superior balance between accuracy and computational efficiency for Discomycetes classification. XAI techniques ensure reliable, interpretable results suitable for biodiversity conservation and ecological research applications.
  • Published in:Biology (Basel),
  • Study Type:Research Study,
  • Source: PMID: 40563969
Scroll to Top