AutoML Applications for Bacilli Recognition by Taxonomic Characteristics Determination over Microscopic Images

Abstract

In this work, we describe our research aimed at developing classifiers for microbial images (bacilli images) obtained through microscopy of live (non-static) samples. We employed our proposed approach called AutoML, which is based on the automatic generation and analysis of the feature space to create the most optimal descriptors for microscopic images used in their classification. This approach allows us to utilize interpretable taxonomic features based on the external geometric characteristics of images of various types of microorganisms. To demonstrate the effectiveness of our proposed solution, we also publish an annotated dataset we collected, containing microbial images of unfixed microscopic scenes. Additionally, we compare the classification performance of our solution with the results of various types of classifiers, including those based on deep neural network models. Our approach showed the best results among those studied (Precision = 0.989, Recall = 0.992, F1-score = 0.990).