Vegetation Reflectance Estimation Through Optimization of a Spectral Unmixing Model
| dc.contributor.advisor | Noble, Scott | |
| dc.contributor.committeeMember | Johnston, James | |
| dc.contributor.committeeMember | Guo, Xulin | |
| dc.creator | Del Castillo Valenzuela, Mikael Fausto | |
| dc.creator.orcid | 0000-0001-6584-4026 | |
| dc.date.accessioned | 2025-05-08T21:12:07Z | |
| dc.date.available | 2025-05-08T21:12:07Z | |
| dc.date.copyright | 2025 | |
| dc.date.created | 2025-04 | |
| dc.date.issued | 2025-05-08 | |
| dc.date.submitted | April 2025 | |
| dc.date.updated | 2025-05-08T21:12:07Z | |
| dc.description.abstract | Spectroscopy has become an extensively used tool for non-destructive high-throughput analysis of plant characteristics such as crop health, stress response, and photosynthetic properties. Hyperspectral cameras and compact spectrometers are common instruments used in agriculture research and field applications to capture the vegetation reflectance spectrum at the canopy scale (about one meter from top of the crop). Although hyperspectral imaging (HSI) allows the isolation of the vegetation spectrum by pixel classification, collecting a clean HSI dataset in the field is challenging due to motion artifacts and storage requirements for extensive areas. A compact spectrometer system is a quicker and smaller data size alternative used for high throughput phenotyping applications. However, this instrument produces a mixed reflectance signal at canopy scale as it integrates all the spectral signals of the scene components within the field of view of the sensor. This study explores the possibility of estimating an isolated vegetation reflectance from a mixed signal using abundance estimates and color information of the scene components. An HSI and a multi-instrument setup built with a compact spectrometer, RGB (red-green-blue) and NIR (near infrared) cameras were used to sample data at canopy scale. The traditional spectral unmixing model and a variation that includes a color-matching process were tested in this study as methods to obtain an isolated vegetation reflectance from a mixed signal. Both spectral unmixing models were evaluated by observing the differences between extracted and estimated endmembers reflectance in a recreated dataset derived from HSI data that represents the multi- instrument setup. The model with color matching showed less estimation error and less sensitivity to noisy inputs in the recreated dataset, and exhibited similar response patterns between the multi-instrument and recreated datasets. This indicates the potential of the proposed method as an alternative for estimating an isolated vegetation spectrum from a mixed signal in applications at the canopy scale. | |
| dc.format.mimetype | application/pdf | |
| dc.identifier.uri | https://hdl.handle.net/10388/16919 | |
| dc.language.iso | en | |
| dc.subject | Spectral unmixing | |
| dc.subject | Reflectance | |
| dc.title | Vegetation Reflectance Estimation Through Optimization of a Spectral Unmixing Model | |
| dc.type | Thesis | |
| dc.type.material | text | |
| thesis.degree.department | Mechanical Engineering | |
| thesis.degree.discipline | Mechanical Engineering | |
| thesis.degree.grantor | University of Saskatchewan | |
| thesis.degree.level | Masters | |
| thesis.degree.name | Master of Science (M.Sc.) |