Akademik Veri Yönetim Sistemi
TROIA 2nd INTERNATIONAL CONFERENCE ON APPLIED SCIENCES, Çanakkale, Türkiye, 25 - 27 Temmuz 2025, ss.199-212, (Tam Metin Bildiri)
Accurate classification of fungi commonly found in nature at the species level is crucial for ensuring food safety, public health, and ecological balance. In particular, incorrect identification of poisonous and hallucinogenic species can lead to serious health problems. Moreover, the use of bioactive compounds contained in edible and medicinal species in modern medicine and industry makes the accurate diagnosis of these organisms even more critical. In this context, the use of image processing and artificial intelligence-based classification systems is gaining importance day by day in order to overcome the limitations of traditional mycological methods. In this study, various deep learning models were compared for their ability to classify fungus species based solely on image data. Widely used CNN architectures such as EfficientNet, ResNet50, DenseNet121, VGG16, and InceptionV3 were evaluated on the same dataset and their validation performance was analyzed. The accuracy rates of the models were found to be ResNet50 (%88.27), EfficientNet (%88.05), VGG16 (%86.46), DenseNet121 (%80.84), and InceptionV3 (%77.96), respectively. In addition, the discrimination power of the models between classes was interpreted in detail with confusion matrix analyses and accuracy/loss curves. It was observed that some models showed a tendency to mix, especially among visually similar species. The findings show that modern architectures such as ResNet50 and EfficientNet can perform high-accuracy classification and that image-based diagnostic systems can provide reliable solutions in mycological applications. This study provides a solid foundation for future mobile applications or decision support systems and provides an artificial intelligence-supported approach to the field of mycology.