Semi-supervised mold differentiation using typical laboratory results as label data

Abstract

This study applies semi-supervised learning to automate the differen-
tiation of mold colonies, thereby reducing the time and cost associated with air
quality assessments. EfficientNet V2 and Normalization-Free Net (NfNet) were
trained on a dataset of mold colony images, created in a semi-supervised way.
NfNet demonstrated superior performance, particularly on non-padded images,
with explainable AI techniques enhancing interpretability. The models exhibited
generalization capabilities to environmental samples, indicating the potential for
automating mold identification and streamlining air quality monitoring, thereby
reducing manual effort and costs. Future work will focus on refining species han-
dling and integrating the system into laboratory workflows.