On the Task of Classifying Microorganisms for Microscopic Analysis

Abstract

The study of microorganisms with a microscope is a task in demand all over the globe, in thousands of bio-laboratories for very different purposes. It can be a simple scientific study of microbial behavior, as well as the study of water quality for the content of pathogenic bacteria, highly demanded in the sanitary examination of various water basins. These and many other tasks require one common action - classification of the microorganism (correct identification of the species) in order to continue the work performed. This step requires highly qualified personnel, whose assistance may not always be possible due to limitations in both personnel resources and time. Therefore, this paper will consider the possibility of replacing the human (full or partial) to simplify the classification process. In this paper, an algorithm for microorganism classification based on YOLOv5 and Mask-RCNN is proposed. A study has been carried out on existing datasets as well as on the dataset collected by the authors. From the data available in public domain, by applying standard algorithms on data augmentation on microorganism search scenarios on image the algorithms used give good result on both detection and segmentation (later applicable for biomass counting). But this result is possible only if there are similar data format and the required microorganisms in the training sample.