UAV photogrammetry and methods of image analysis in agriculture experiments

The main goal of this diploma thesis was the processing of UAV images of selected agricultural crops using photogrammetric methods and image analysis methods to evaluate the usability of these image data and products derived from them for various purposes and applications in agriculture. The primary data for this study were UAV images taken by the drone DJI Phantom 4 Pro in the 2019 season and data of control points locations measured during the UAV imaging. Another task was to use the products created by processing UAV images for the calculation of canopy height models and derived canopy height metrics and for the calculation of selected vegetation indices based on the transformation of RGB bands. The partial goal of this study was to find a suitable algorithm using image analysis of RGB images for automatic detection of the number of wheat ears. During the processing of this study, there was also a requirement to perform reference imaging together with manual measuring of crop heights at selected reference points in order to compare these heights with the heights derived from the resulting canopy height models. The reference imaging and measuring was done additionally in June 2020. The programs Agisoft Metashape, Cloud Compare, ArcGIS Pro, ImageJ and Microsoft Excel were used for data processing and for obtaining the final products.
The preliminary research was dedicated to the sources of the particular topic. Through the study of literature, articles, previous research and studies, the necessary information about the possibilities of using UAV photogrammetry and image data analysis for research and applications in agriculture was obtained. The theoretical part of the thesis describes basic information about photogrammetry in general, about UAV photogrammetry and the possibilities of its use. Next, selected algorithms and approaches for generating three-dimensional point clouds based on two-dimensional photographs and basic information about canopy height models and derived canopy height metrics were described here. At the end of the theoretical part, the image analysis of data was described, including information about basic image operations and selected vegetation indices.
The next chapter describes the current issues within the topic of the thesis with the introduction of mostly foreign studies that struggled with similar issues or that otherwise helped meet the objectives of this thesis. Before processing the primary data, it was important to obtain a theoretical basis and an overview of the photogrammetric methods and methods of image analysis which are available in the mentioned programs and the possibilities of their usage for the purposes of this study.
Within the practical part of this thesis, first, the parameters of the acquired data and performed image flights were described. This was followed by a description of the procedure used in the processing of the acquired data and products derived from them, which included a manual with a description of the methods, tools and their parameters, including the description of used setting.
The next chapter of the practical part describes specific case studies that were solved in this study. This chapter first presents case studies for the processing of wheat and garlic images taken in 2019. Another case study describes the implementation of reference imaging together with the manual measurement of crop heights in the field. In this case study, the accuracy of the derived heights of the created height models in comparison with the focused heights is evaluated from the processed data using reference points. The last case study is focused on finding and testing an algorithm suitable for automatic detection of the number of wheat ears with using UAV images taken in the visible spectrum. By using three test areas, in which the ears were counted manually by the author of this diploma thesis, the optimal setting of individual tools of the algorithm was searched and next step was an evaluation of the accuracy of the resulting numbers of detected ears in comparison with the numbers get with manual counting. In the second part of this case study, the algorithm with the found optimal settings was tested on the created orthophoto image.
The results of this diploma thesis showed that with the help of UAV images taken in the visible spectrum, dense point clouds can be generated, which can be further used to derive products usable for research and applications in agriculture. Although it was not possible to construct accurate canopy height models from the created point clouds, which would correspond to the measured heights of the modeled agricultural crops, these models can be used to compare crop heights within the modeled area. The results further showed that UAV images taken in the visible spectrum can be used with using the methods of image analysis for automatic wheat ears detection and counting. Although the algorithm for automatic detection of wheat ears did not detect the exact numbers of ears on test wheat images, which were manually counted, the detected numbers were very close (success rate> 90%) to the numbers obtained by manual counting. The resulting detected numbers of ears are therefore at least mutually comparable and with their help it is possible to determine in which parts of the investigated fields the density of ears is higher or lower. The results of the correlation matrix in the case of wheat have shown that there may be correlations and dependencies between certain derived height metrics and calculated vegetation indices. The outputs and results of this diploma thesis may be useful for future research of agricultural crops or for other applications in agriculture.
Finally, the text part of the work was written. The whole work, including the text part, appendices, outputs, source and created data is available in appendix 8 on DVD. The results of the thesis are also presented on the website of the diploma thesis and the information poster.