Crimaldi, Mariano (2022) VISmaF: immersive virtual visualization in smart farming and digital advances in agriculture. [Tesi di dottorato]

Preview

Text Crimaldi_Mariano_34.pdf Download (39MB) | Preview

Collection description

Digital technologies have taken hold over the past few decades in many scientific and industrial sectors: from the aviation industry to medicine, from engineering to agriculture. Thanks to the rapid development of hardware and software solutions, it has been possible to implement techniques and technologies that were unthinkable until recently. A striking example is the adoption of the Digital Twin (DT), which is a digital counterpart, perfectly reproduced, of the object or model that is to be studied. The adoption of a DT allows the study, through numerical models, of complex systems without having to materially build, implement or in some cases destroy them, simplifying the search for new solutions also from an economic point of view. The digitization of processes and their scientific study has also been introduced in agriculture where today it is possible to adopt advanced systems for remote sensing, management of fields or farms, decision support. In the specific case of crop modeling, great advances have been made in the last decade thanks to the development of the so-called FSPM, or biological/mathematical models that take into account the functional processes of a plant, in connection and in relation to its geometric structure. These models allow the creation of a particular DT called plant in silico that is a digital plant with which it is possible to understand, study and visualize all the possible changes made at physiological, genetic, phenological and other levels. In silico approaches can also be used at a larger scale, at the stand level or on a territorial scale, to study broader phenomena. Also at the stand or spatial level, the introduction of remote sensing in agriculture has allowed further digitization of the field. The introduction in agriculture in recent years of increasingly accessible and easy-to-use systems, thanks to sensors mounted on Unmanned Aerial Vehicles, has allowed frequent, robust and accurate data collection. These data are used in a variety of ways: from creating vegetation indices to obtain prescription maps, to using them in Deep Learning algorithms for automatic recognition of diseases, phenological status, land uses and more. The present study aimed to develop advanced digitization techniques for some applications in the field of agriculture with the goal of presenting and deploy some products used in concrete applications by the industry of the sector. The state of the art of FSPM models has been reviewed, in particular those with a 3D tree structure output. This review showed a lack of system-based models so a proof of concept of a biological/mathematical system-dynamics model coupled with a real-time 3D rendering engine has been proposed. This solution has been studied because there is a rapidly expanding industry sector that is, as a particular DT, process virtualization. In agriculture, process virtualization, unlike the process industry, is mostly absent. The study therefore proposes itself as a possible model of virtualization of biologically verisimilar trees (the aforementioned in silico plants) in order to be able to use them in industrial agronomic applications, such as training of agricultural operators, without having to physically grow, modify or perhaps destroy a real tree. This proof of concept has been then developed in detail, developed with its characteristics, its code, its implementation and its possible agronomic applications. A Deep Learning system for weed recognition has been developed in chapter in this study, carried out in order to have an automated aerial (UAV) or terrestrial (UTV) system that allows selective recognition of weeds to be removed and not to be removed in order to provide farmers or the industry with a Site-Specific Weed Management tool that allows them to save on herbicide use. Also, a study has been developed on the interaction between water flows and plants in vegetated channels in addition to other digital agriculture applications for eco-hydraulics applications. The study has been possible thanks to the adoption of some digital techniques such as 3D scanning of plants to create a DT used in numerical finite element simulations. In addition, Remote Sensing approaches have been developed using multispectral sensors on drones to calculate the Leaf Area Index (LAI) in hard-to-access vegetated channels and assess the vegetation cover of the channel. This study on vegetated channels has been carried out because the results obtained can be used by the mechanical, aerospace or naval industry to produce vehicles optimized to work in conditions of high turbulence, or more in a smaller sense to manage vegetated channels to optimize hydraulic flows without compromising the plant ecosystem. The projects presented in this work show how the digitization of agriculture is a broad field, both in terms of applications and theory. The interest shown by industry in these applications raises hopes for rapid future development in what are the different fields of application.

Downloads

Actions (login required)

View Item