Microbial Image Classification with Limited Data Quality: A Progressive Enhancement-Fusion Approach

Microbial classification holds significant application value in biomedical testing and environmental monitoring. However, conventional microscopic identification methods suffer from limitations including low efficiency and high technical expertise requirements. To address challenges such as low resolution and high noise in microscopic images, this study proposes a novel classification method integrating data augmentation with multi-model ensemble: The AugMix algorithm is used to generate enhanced samples to improve the robustness of the model to image distortion; A hybrid architecture is constructed to integrate the advantages of each model through a probability layer fusion strategy; it aims to provide a reliable automatic classification solution for low-quality microbial images. Experiments on 8 types of microbial microscopic data sets (including 2,400 images of amoeba, bacteria, etc.) show that this method significantly improves the classification accuracy.