Light microscopic analysis of diatom frustules is widely used both in basic and applied research, notably taxonomy, morphometrics, water quality monitoring and paleo-environmental studies. In these applications, usually large numbers of frustules need to be identified and/or measured. Although there is a need for automation in these applications, and image processing and analysis methods supporting these tasks have previously been developed, they did not become widespread in diatom analysis. While methodological reports for a wide variety of methods for image segmentation, diatom identification and feature extraction are available, no single implementation combining a subset of these into a readily applicable workflow accessible to diatomists exists.
Deep convolutional neural networks are emerging as the state of the art method for supervised classification of images also in the context of taxonomic identification. Different morphologies and imaging technologies applied across organismal groups lead to highly specific image domains, which need customization of deep learning solutions. Here we provide an example using deep convolutional neural networks (CNNs) for taxonomic identification of the morphologically diverse microalgal group of diatoms. Using a combination of high-resolution slide scanning microscopy, web-based collaborative image annotation and diatom-tailored image analysis, we assembled a diatom image database from two Southern Ocean expeditions. We use these data to investigate the effect of CNN architecture, background masking, data set size and possible concept drift upon image classification performance. Surprisingly, VGG16, a relatively old network architecture, showed the best performance and generalizing ability on our images. Different from a previous study, we found that background masking slightly improved performance. In general, training only a classifier on top of convolutional layers pre-trained on extensive, but not domain-specific image data showed surprisingly high performance (F1 scores around 97%) with already relatively few (100–300) examples per class, indicating that domain adaptation to a novel taxonomic group can be feasible with a limited investment of effort.
Light microscopy analysis of diatom frustules is widely used in basic and applied research, notably taxonomy, morphometrics, water quality monitoring and paleo-environmental studies. Although there is a need for automation in these applications, various developments in image processing and analysis methodology supporting these tasks have not become widespread in diatom-based analyses. We have addressed this issue by combining our automated diatom image analysis software SHERPA with a commercial slide-scanning microscope. The resulting workflow enables mass-analyses of a broad range of morphometric features from individual frustules mounted on permanent slides. Extensive automation and internal quality control of the results helps to minimize user intervention, but care was taken to allow the user to stay in control of the most critical steps (exact segmentation of valve outlines and selection of objects of interest) using interactive functions for reviewing and revising results. In this contribution, we describe our workflow and give an overview of factors critical for success, ranging from preparation and mounting through slide scanning and autofocus finding to final morphometric data extraction. To demonstrate the usability of our methods we finally provide an example application by analysing Fragilariopsis kerguelensis valves originating from a sediment core, which substantially extends the size range reported in the literature.