Inovações tecnológicas baseadas em ferramentas de baixo custo para imageamento de organismos do solo assistido por visão computacional

Abstract

The machinery and analysis used in soil organisms studies have been in constant evolution. Usually, microscopes are used to carry out some of those analyses. Though commonly employed in labs, this equipment is usually expensive. One of the components which could be important to the elevated cost of a microscope is the automatized stage. That cost can be up to dozens of thousands of dollars. One strategy that can be used to overcome that barrier is acquiring the stage within the scope of Open Science. With that approach, researchers can assemble their own equipaments using softwares and hardwares of open source. That approach will reduce costs drastically to acquire an automatized stage (which will increase the analysis capacity when compared to traditional microscopes). Moreover, the automatized stage can be adapted to local conditions (for example parts of an old microscope can be reutilized, then reducing costs), as well as to sample’s particularities. This versatility is fundamental in plant parasites nematodes. Traditionally those organisms have been observed in non-conventional microscope slides (wider and longer than conventional ones). That reality makes the observation of nematodes under a microscope more difficult, because the stages of microscopes usually are smaller than the slides dimensions. We have to bear in mind that one of the main uses we can employ to an automatized stage is scanning samples automatically (that means, capturing images from the totality of the sample). In this approach, multiple images (which represent the totality of the sample or only a region) are acquired. It is also necessary a strategy to stitch those images, so that, in the end, the visual field can be amplified by combining multiple images into a single one. Besides that, methods used in soil organisms analyses are in constant change. Recently, machine learning tools have been utilized in the study of a variety of organisms such as bacteria, fungi, and nematodes. This approach is advantageous, particularly in analyses that are labour-intensive and/or need a specialized worker. In that scenario, trained (and tested) algorithms can be deployed to perform those tasks. Aiming to contribute to the popularization of automatized stages and of computer vision (a sub-area of machine learning) approaches to soil organisms analyses, we present in this dissertation two low-cost automatized stages (the first one reuses an old bright-field microscope, while the second one is based on a low-cost digital microscope that has enough resolution to observe nematodes in parasitized soybean roots). Adding to that, we present here algorithms that make some measures on nematodes images, select images with nematodes to posterior stitching to create images of regions containing those organisms in order to amplify the visual field over the sample, and an algorithm to perform segmentation of arbuscular mycorrhizal fungi’s hyphae. To write both softwares of control stages, we utilized the programming language Python. For the hyphae’s segmentation algorithm we also used a Python code. It was obtained satisfactory results in imaging mostly regions containing nematodes with both high resolution and wide vision field, as well as for the measurements of nematodes in images, and hyphae segmentation. Altogether, we believe, it is presented here a cost-effective approach that can popularize the use of automatized stages and machine learning tools for the study of soil organisms.