Automation of Diplococci Recognition in Microscope Images

In this study, we outline our efforts to create classifiers for microbial images, particularly targeting diplococci images captured through microscopy of live (dynamic) samples. Our approach, named AutoML, involves the automatic creation and analysis of feature spaces to develop the most effective descriptors for classifying microscopic images. This methodology allows the use of interpretable taxonomic features derived from the external geometric properties of various microorganisms' images. To validate our method, we have published an annotated dataset we gathered, which includes microbial images of non-fixed microscopic scenes. Additionally, we compared the classification performance of our technique with various other classifiers, including those based on deep neural networks. Our approach achieved the highest performance among those tested, with a Precision of 0.993, Recall of 0.990, and an F1-score of 0.991.