Abstract

Fungal contamination on space vessels can be of great risk to the health of astronauts and the durability of the vessels. The manual microscopic classification is both time-consuming and non-feasible in space missions, and thus it is essential to automate the process. The study involves the model development of three deep learning models, namely the baseline CNN, EfficientNetB0, and MobileNetV2, for the classification of five fungal species using a secondary DeFungi microscopic image database. After image augmentation and common preprocessing, each model was tested in terms of precision, recall, F1-score, and ROC-AUC. Findings indicate that the MobileNetV2 model performed better than the other models, which registered 67% test set accuracy and excellent performance across most classes. It is compact and lightweight and can make onboard fungal monitoring easy in resource-limited settings. In this study, the research is used to indicate how to achieve autonomous bio-surveillance advancement through the use of deep learning in long-duration space missions.

Statistics