Theses and Dissertations @ UNI

Availability

Thesis (UNI Access Only)

Abstract

Obtaining site-specific knowledge of variables affecting crop productivity is a critical step in precise agricultural management. Traditional map-based approaches where soil sampling and subsequent laboratory analysis are used to create variability maps have proven useful, but remote-sensing approaches are more desirable when cost and accuracy limitations can be overcome. Small unmanned aircraft capable of carrying imaging devices have become more accessible and increasingly autonomous over the last ten years. Also, hyperspectral imagers capable of providing high spectral resolution data are being optimized for use on light-weight aerial platforms. These platform and payload advances pose new opportunity for wide scale, high resolution data collection of agricultural lands. However, processing chains and analytical methods to compliment the availability of these data are lacking. This study develops a processing chain for framebased hyperspectral data and assesses whether genetically distinct maize varieties can be automatically separated and classified using various analytical methods. Our field work included UAV data collection campaigns completed on five sampling dates during July 2014 (02, 07, 10, 21, 24) in Scott County, Iowa. The site contained 13 genetically unique varieties of maize planted adjacent to one another on 30 May, 2014. Using 24 bands of the hyperspectral payload as response variables, we found each hybrid to have a statistically significant difference from one another on all five sampling dates. Discriminant analysis and SIMCA classification showed general agreement in ability to separate samples based on building models of the hybrid classes from known observations. The global model accuracies for SIMCA analyses on each date were 32%, 28%, 28%, 32% and 21%, respectively. Discriminant analysis successful classification rates for days ranged from 32% during sampling period 5 to almost 38% within sampling period 4. We conclude subtle differences in crops expressed as variations in spectral signatures, such as those imposed by mildly different genetic characteristics, are possible to detect using these new hardware and processing chains. However, reliably classifying the independent variables based solely on these subtle spectral differences is not possible using these methods.

Year of Submission

2019

Degree Name

Master of Arts

Department

Department of Geography

First Advisor

Dr. Patrick Pease, Chair, Thesis Committee

Date Original

2019

Object Description

1 PDF ( VIII, 101 pages)

Language

en

Off-Campus Download

Share

COinS