| dc.contributor.author | Benson, Albert Ansah | |
| dc.date.accessioned | 2026-11-13T22:21:32Z | |
| dc.date.graduationmonth | December | |
| dc.date.issued | 2026 | |
| dc.description.abstract | Nitrogen fertilizer is one of the largest variable input costs in maize production, yet much of the nitrogen applied is not taken up by the crop and is lost through leaching, denitrification, volatilization, and runoff, at a cost to both the farmer and water quality. Because crop nitrogen need varies across a field and from season to season, a single rate decided at planting is often poorly matched to it. A rapid in-season measure of tissue nitrate-N would allow fertilizer to be applied where and when the crop needs it, while a sidedress application can still be made. Healthy green leaves have a distinctive optical response: chlorophyll absorbs much of the red light that reaches the leaf, whereas internal leaf structure responds strongly in the near-infrared part of the electromagnetic spectrum. Multispectral sensors use this contrast to detect changes in crop condition. However, common vegetation indices often reduce the response to two or three wavelengths, and their interpretation can be affected by crop development, ambient light, and leaf-surface reflection. This study therefore developed a handheld active optical sensor that expanded the measurement to six wavelengths and evaluated its ability to estimate tissue nitrate-N as an indicator of nitrogen status in maize. The sensor sequentially illuminated maize leaves at 660, 670, 700, 750, 780, and 850 nm. A silicon photomultiplier detected the reflected light, which was amplified and digitized with a 16-bit analog-to-digital converter. Measurements of a 99% reflectance Spectralon reference confirmed a measurable response at all six wavelengths. The instrument was then evaluated in a randomized complete block field experiment with six nitrogen treatments, three blocks, and two planting dates. At V10, R1, and R3, spectral readings were collected from individual leaves under outdoor ambient light and inside a dark enclosure. Tissue nitrate-N determined in the laboratory served as plant-level ground truth. Under controlled illumination, the sensor recovered the expected lower red and higher near-infrared responses of healthy green leaves. Nevertheless, spectral profiles and NDVI overlapped substantially among nitrogen treatments, so the optical measurements did not consistently separate treatment levels. A senesced-leaf test produced a stronger red response and much lower NDVI than healthy green leaves. Thus, the sensor detected a large change in leaf optical condition, although this did not demonstrate sensitivity to the smaller physiological differences among healthy plants. Tissue nitrate-N ranged from 10 to 13,937 ppm and was strongly right-skewed; therefore, it was modeled as log(1 + nitrate-N) variable. The data were divided by plot into 174 training plants and 59 test plants. Because growth stage affected both nitrate-N and spectral response, growth stage was included as contextual information. A feed-forward neural network using the six wavelengths and growth stage achieved R² = 0.752 and r = 0.902 on the transformed scale in the dark enclosure. Outdoors, performance decreased to R² = 0.526 and r = 0.810. Withholding growth stage reduced R² to 0.471 in the dark enclosure and 0.096 outdoors, confirming that developmental context was especially important. Individual wavelengths carried little predictive information, whereas performance improved as wavelengths were combined. When the six wavelengths were replaced by nine NDVI combinations formed from the same six wavelengths, the dark-enclosure model reached R² = 0.542 and r = 0.771, and the outdoor model reached R² = 0.130 and r = 0.470, which indicates that reducing the bands to NDVI discarded information useful for estimating nitrate-N. Finally, a controlled angle experiment with six leaves showed that leaf orientation significantly affected the response at all six wavelengths and suggested a small, wavelength-dependent specular component, largest in the red region, although this attribution could not be confirmed and its contribution to nitrate-N prediction error could not be quantified. Overall, the prototype provides a functional platform for controlled, relative measurements of maize leaf reflectance when the six-wavelength response is interpreted with growth stage. The results do not establish an absolute nitrate-N calibration or support field-ready fertilizer recommendations. Further work would improve ambient-light exclusion, use cross-polarization to suppress surface reflection, validate performance across seasons and locations, and collect chlorophyll and total leaf nitrogen measurements, which should help improve the sensor performance | |
| dc.description.advisor | Naiqian Zhang | |
| dc.description.degree | Master of Science | |
| dc.description.department | Department of Biological & Agricultural Engineering | |
| dc.description.level | Masters | |
| dc.description.sponsorship | United States Department of Agriculture (USDA) National Institute of Food and Agriculture (NIFA), Award Number 2023-67021-44838 | |
| dc.identifier.uri | https://hdl.handle.net/2097/47509 | |
| dc.language.iso | en_US | |
| dc.subject | Maize | |
| dc.subject | Nitrate | |
| dc.subject | Active optical sensor | |
| dc.subject | Reflectance spectroscopy | |
| dc.subject | Nitrogen management | |
| dc.subject | Machine learning | |
| dc.title | A handheld optical sensing device for assessing nitrate concentration in maize leaves using reflectance spectroscopy | |
| dc.type | Thesis |
English
العربية
বাংলা
Català
Čeština
Deutsch
Ελληνικά
Español
فارسی
Suomi
Français
Gàidhlig
ગુજરાતી
हिंदी
Magyar
Italiano
Қазақ
Latviešu
मराठी
Nederlands
Polski
Português
Português do Brasil
Русский
Srpski (lat)
Српски
Svenska
தமிழ்
Türkçe
Yкраї́нська
Tiếng Việt
繁体中文