Deep Ensemble Learning and Explainable AI for Multi-Class Classification of Earthstar Fungal Species

Author: mycolabadmin
9/23/2025
View Source

Summary

Scientists created an artificial intelligence system that can identify eight types of earthstar mushrooms from photographs with over 96% accuracy. These mushrooms look very similar to each other, making them difficult to tell apart by eye alone. The AI system not only identifies the mushrooms correctly but also shows which parts of the image it looked at to make its decision, making it transparent and trustworthy. This technology could help scientists monitor wild mushroom populations and improve sustainable farming practices.

Background

Earthstar fungi are morphologically similar ectomycorrhizal species that are frequently misidentified by traditional methods. Artificial intelligence-based image processing offers potential for accurate automated classification of these fungi. This study addresses the challenge of distinguishing eight visually similar Earthstar species that have never been analyzed together as a group.

Objective

To develop and evaluate deep learning models for multi-class classification of eight Earthstar fungal species (Astraeus hygrometricus, Geastrum coronatum, G. elegans, G. fimbriatum, G. quadrifidum, G. rufescens, G. triplex, and Myriostoma coliforme) and to enhance model interpretability through explainable AI techniques. The study aimed to compare CNN and transformer architectures and develop ensemble models that balance classification accuracy with explainability.

Results

EfficientNet-B3 achieved the best individual performance at 96.23% accuracy with 0.9570 MCC and 0.1050 log loss. The EfficientNet-B3 + DeiT ensemble demonstrated strong balanced performance with 93.83% precision, 93.72% recall, 93.73% F1-score, and 0.9282 MCC. Ensemble models reduced confusion between morphologically similar species, particularly Geastrum triplex and G. fimbriatum. Explainability analysis revealed biologically meaningful attribution patterns focused on discriminative features like the spore sac and peristome.

Conclusion

The study successfully demonstrates that deep learning with ensemble strategies can classify morphologically similar Earthstar fungi with high accuracy (>96%) while maintaining interpretability. The integration of explainable AI techniques provides transparent decision-making that is valuable for scientific applications. This framework has practical applications for fungal monitoring in agricultural ecosystems and sustainable production strategies, with potential for expansion to diverse ecological regions and field conditions.

Published in:Biology (Basel),
Study Type:Experimental Study,
Source: PMID: 41154716, DOI: 10.3390/biology14101313