Andmebaasi logo

3. Doktoritööd

Selle kollektsiooni püsiv URIhttp://hdl.handle.net/10492/2490

Sirvi

Sirvi 3. Doktoritööd Märksõna "agricultural technology" järgi

Nüüd näidatakse 1 - 1 1
  • Pisipilt
    Kirje
    Unmanned aircraft systems and image analysis in yield estimation and agricultural management
    (Estonian University of Life Sciences, 2022) Li, Kai-Yun; Sepp, Kalev; Vain, Ants; Burnside, Niall; Institute of Agricultural and Environmental Sciences; Tsai, Hui-Ping (opponent)
    This thesis aims to examine how machine learning (ML) technologies have aided significant advancements in image analysis in the area of precision agriculture. These multimodal computing technologies extend the use of machine learning to a broader spectrum of data collecting and selection for the advancement of agricultural practices (Nawar et al., 2017) These techniques will assist complicated cropping systems with more informed decisions with less human intervention, and provide a scalable framework for incorporating expert knowledge of the PA system. (Chlingaryan et al., 2018). Complexity, on the other hand, can be seen as a disadvantage in crop trials, as machine learning models require training/testing databases, limited areas with insignificant sampling sizes, time and space-specificity, and environmental factor interventions, all of which complicate parameter selection and make using a single empirical model for an entire region impractical. During the early stages of writing this thesis, we used a relatively traditional machine learning method to address the regression problem of crop yield and biomass prediction [(i.e., random forest regression (RFR), support vector regression (SVR), and artificial neural network (ANN)] to predicted dry matter (DM) yields of red clover. It obtained favourable results, however, the choosing of hyperparameters, the lengthy algorithms selection process, data cleaning, and redundant collinearity issues significantly limited the way of the machine learning application. We will further discuss the recent trend of automated machine learning (AutoML) that has been driving further significant technological innovation in the application of artificial intelligence from its automated algorithm selection and hyperparameter optimization of the deployable pipeline model for unravelling substance problems. However, a present knowledge gap exists in the integration of machine learning (ML) technology with unmanned aerial systems (UAS) and hyperspectral-based imaging data categorization and regression applications. In this thesis, we explored a state-of-the-art (SOTA) and entirely open-source AutoML framework, Auto-sklearn, which was built on one of the most frequently used machine learning systems, Scikit-learn. It was integrated with two unique AutoML visualization tools to examine the recognition and acceptance of multispectral vegetation indices (VI) data collected from UAS and hyperspectral narrow-band VIs across a varied spectrum of agricultural management practices (AMP). These procedures incorporate soil tillage method (STM), cultivation method (CM), and manure application (MA), and are classified as four-crop combination fields (i.e., red clover-grass mixture, spring wheat, pea-oat mixture, and spring barley). Additionally, they have not been thoroughly evaluated and lack characteristics that are accessible in agriculture remote sensing applications. This thesis further explores the existing gaps in the knowledge base for several critical crop categories and cultivation management methods referring to biomass and yield analysis, as well as to gain a better understanding of the potential for remotely sensed solutions to field-based and multifunctional platforms to meet precision agriculture demands. To overcome these knowledge gaps, this research introduces a rapid, non-destructive, and low-cost framework for field-based biomass and grain yield modelling, as well as the identification of agricultural management practices. The results may aid agronomists and farmers in establishing more accurate agricultural methods and in monitoring environmental conditions more effectively.