Advanced deep learning approaches for the accurate classification of Phallaceae fungi with explainable AI

Abstract

In this study, deep learning (DL)-based models were developed for the classification of 5 fungal species from the Phallaceae family (Clathrus ruber, Colus hirudinosus, Mutinus caninus, Phallus impudicus, and Pseudocolus fusiformis). ConvNeXT achieved the highest performance with 98% accuracy, 98% precision, 98% recall, and 98% F1-score. EfficientNetB4 and Xception also performed well with 96% accuracy. In contrast, lighter models such as MobileNetV2 and MixNet S showed significantly lower accuracy (84% and 80%, respectively). Among the explainable artificial intelligence (XAI) techniques, gradient-weighted class activation mapping (Grad-CAM) and Integrated Gradients showed that high-accuracy models focus more effectively on biologically meaningful regions. In particular, the ConvNeXT plus Grad-CAM combination consistently highlighted critical structural areas, such as the cap and stalk of fungi, resulting in more accurate classifications. These findings show that DL-based models offer high accuracy in classifying fungal species with complex morphological features. Furthermore, XAI techniques play a critical role in enhancing classification processes.