Skip to main content
ARS Home » Research » Publications at this Location » Publication #130425
Title: SPECTRAL MIXING APPROACH TO IMAGE ANALYSIS FOR PRECISION AGRICULTURE

Author
item Fitzgerald, Glenn
item Barnes, Edward
item Pinter Jr, Paul
item Clarke, Thomas

Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: 3/4/2002
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Remote sensing has been used in precision agriculture to relate variation in the imagery to measured ground conditions such as crop nutrient status or canopy density. Generally, the raw image data are transformed to derived values that relate the spectral domain to biophysical parameters, such as correlating vegetation indices like the Normalized Difference Vegetation Index (NDVI) to percent cover or categorizing image variation using supervised and unsupervised classifications. A technique new to precision agriculture, called spectral mixture analysis (SMA), may be able to improve the derived relationships between imagery and measured ground data by providing fraction maps of the parameters of interest. This has the potential to quantify the relative amount of a component present and locate it within a field. Remotely-sensed imagery was acquired on six dates in 2001 from a 3.4-ha cotton field at the Maricopa Agricultural Center, Maricopa, Arizona. Sensors included a thermal scanner (8-14 microns), a color digital camera, and a Dycam agricultural camera that provided red and near-infrared images. All images were georegistered to known ground locations visible in the images. Various image maps were produced based on different image analysis techniques including NDVI, supervised classification, unsupervised classification, and fraction maps derived from SMA. Locations were selected in the imagery where various crop and soil parameters were collected. Ground data collected within 48 hours after flights included percent crop cover, plant height, leaf area index, total plant fresh weight, dry weight of plant components, petiole nitrate content, and soil texture. Mean pixel values were calculated