研究内容

Filamentous fungi are known to cause various issues such as toxicity, allergies, and diseases across diverse fields including residential environments, food production, and healthcare. To mitigate these risks, rapid and simplified identification of filamentous fungi is essential. However, accurate identification typically requires DNA analysis, which is both time-consuming and costly. On the other hand, conventional simplified methods rely heavily on the advanced expertise of trained specialists.
To address these challenges, this research aims to develop an image-based diagnostic method utilizing machine learning, enabling anyone to easily and quickly identify filamentous fungi. Given the significant effort required for labeling training data, we apply techniques such as transfer learning, data augmentation, and semi-supervised learning to achieve high-accuracy identification even with limited datasets. Furthermore, to ensure the method can be deployed on small-scale systems, we are exploring the use of knowledge distillation techniques developed in our previous studies.
This project contributes to the advancement of accessible and efficient fungal diagnostics, with potential applications in public health, food safety, and environmental monitoring.