Jenal, Alexander Gerhard Josef ORCID: 0000-0002-1890-4839 (2022). A UAV-BASED VNIR/SWIR MULTISPECTRAL MULTI-CAMERA SYSTEM FOR MONITORING VEGETATION - Development, Application and Evaluation on Agricultural Systems. PhD thesis, Universität zu Köln.
|
PDF
Dissertation_final_AJ_print.pdf Download (104MB) | Preview |
|
PDF (Unformatierte Textdatei)
Abstract_MetaData_Diss.txt - Additional Metadata Restricted to Repository staff only Download (15kB) |
Abstract
A growing world population with limited or even declining resources and intensifying climate change are putting increasing pressure on global food production and, in particular, on the agricultural sector. Driven by technological innovations, precision agriculture, and smart farming are upcoming management strategies essential for counteracting these negative impacts. Both trends aim to use site-specific management practices to selectively account for multifaceted plant variability within croplands to maximize yields with minimal inputs, which should also lead to more sustainable agricultural management practices. However, since one cannot manage what one cannot measure, timely information on current in-field conditions is of the utmost importance for effectively implementing these agronomic management strategies. In this regard, optical remote sensing systems based on unmanned aerial vehicles (UAVs) can significantly contribute to airborne spectral vegetation analysis of small and medium-sized agricultural areas, both in research and in agricultural applications. Specifically, multispectral multi-camera systems are often used to acquire spectral image data sets for deriving established vegetation indices (VIs), which are then used as the basis for non-destructive estimation of specific crop traits to support—for example, site-specific crop management decisions in precision agriculture. These imaging systems can be effortlessly integrated into various UAV systems, and the spectral image data obtained has a high spatial resolution. Furthermore, these acquired data sets can be processed in photogrammetric workflows to derive spectral orthomosaics and 3D structural information. However, these commercial or experimental systems consist exclusively of multiple silicon-based image sensors. Moreover, they are in most cases self-contained and not adaptable to different applications, thus limiting their spectral resolution to a small number of primarily broad spectral bands in the visible (VIS) and near-infrared (NIR) domain (400 to 1000 nm). However, these multi-camera systems do not cover the short-wave infrared (SWIR) range from 1000 to 1700 nm, which contains crucial spectral absorption features related to biomass, water, and biochemical leaf constituents such as lignin, starch, and nitrogen (N). Furthermore, as observed in hyperspectral imaging applications, there is already a need for narrowband SWIR imaging, especially with the more feasible UAV-based multispectral multi-camera systems. Their application in agricultural systems, with adapted spectral sensitivity in the SWIR, would be ideal for detecting site-specific, subtle spectral vegetation features in high spatial resolution. Similarly, for multispectral Earth observation satellite missions, the SWIR has minimal, still very broadband, coverage with spatial resolutions unsuitable for site-specific applications or phenotyping, resulting in a spatial and spectral sampling gap in this area. Similarly, hyperspectral missions that could compensate, at least, for this spectral shortfall, are currently only on a limited scale operational and also have lower ground resolution. Thus, to detect light of the (SWIR) domain, a different sensor type—namely, one based on the semiconductor material InGaAs—is required. However, until recently, these sensors were not performant enough to be incorporated into miniaturized camera modules. In addition, due to technical limitations, InGaAs sensors were comparatively expensive compared to their silicon counterparts. However, these shortfalls have improved and therefore, this research (1) developed a UAV-based 2D imaging prototype of a narrowband two-channel multispectral multi-camera system, based on InGaAs sensors, with a spectral response from VIS to SWIR (approximately 600 nm to 1700 nm) and (2) evaluated the potential of this camera system prototype for estimating crop traits of two different agricultural vegetation types from the acquired spectral image data sets. These outcomes were accomplished through the objectives pursued in the context of three peer-reviewed studies published in scientific journals. The first study focused on the development process, including selecting dedicated parts for the optical components, suitable InGaAs camera sensor modules, and a powerful computing unit. Interchangeable filter assemblies have been developed to acquire a freely selectable narrow wavelength band with each integrated sensor. Two options for this critical part were implemented and tested for suitability to be operated in the system’s optical path. Subsequently, several custom mechanical and electronic components were designed and fabricated to build the integrated multi-camera system. Finally, successful tests of the individual components and the overall system in a climate chamber and an integrating sphere in a spectral laboratory, as well as an initial test flight, completed the overall development process. The subsequent two case studies evaluated the potential of the newly developed SWIR-based multi-camera system for estimating vegetation traits in a permanent grassland field trial and a winter wheat field trial. For this purpose, two established SWIR-based VIs were selected: the two-band normalized ratio index (NRI) and the four-band GnyLi index. The data collection took place in one flight campaign for each experimental field to acquire four spectral image datasets per trial with two UAV flights in direct succession. The resulting reflectance data sets were processed into orthomosaics for each wavelength band. The GnyLi and NRI VIs were then derived from these orthomosaics and further analyzed for their estimation accuracy in bivariate regression models using trial-specific spectral and destructively sampled ground truth data. In summary, it was concluded that the findings of both case studies confirmed the camera system’s presumed performance from the initial tests in the first study. The overall aim of developing and evaluating the narrowband VNIR/SWIR multispectral multi-camera system for UAV-based vegetation monitoring was thus successfully achieved. The derived VIs from spectral image data acquired by the camera system reached high accuracies in estimating dry matter yield in permanent grassland and fresh biomass, dry biomass, moisture content, N concentration, and N uptake in winter wheat. Further multi-temporal and multi-annual investigations with refined analysis methods should be carried out to confirm the findings in this thesis. As a result, this more in-depth research could further evaluate and optimize the SWIR-based multi-camera system for potential use in application-specific UAV-based monitoring applications in the context of precision agriculture.
Item Type: | Thesis (PhD thesis) | ||||||||||
Translated abstract: |
|
||||||||||
Creators: |
|
||||||||||
URN: | urn:nbn:de:hbz:38-616888 | ||||||||||
Date: | 2022 | ||||||||||
Language: | English | ||||||||||
Faculty: | Faculty of Mathematics and Natural Sciences | ||||||||||
Divisions: | Faculty of Mathematics and Natural Sciences > Department of Geosciences > Geographisches Institut | ||||||||||
Subjects: | Generalities, Science Data processing Computer science Education Natural sciences and mathematics Physics Earth sciences Technology (Applied sciences) Agriculture |
||||||||||
Uncontrolled Keywords: |
|
||||||||||
Date of oral exam: | 4 May 2022 | ||||||||||
Referee: |
|
||||||||||
Refereed: | Yes | ||||||||||
URI: | http://kups.ub.uni-koeln.de/id/eprint/61688 |
Downloads
Downloads per month over past year
Export
Actions (login required)
View Item |