Dense Time Stack Change Detection Analysis of Cropland Phenology Using Images from Unmanned Aerial Systems

Abstract

Unmanned Aerial Vehicles (UAVs) offer a unique alternative to traditional satellite based remote sensing techniques. High-resolution imagery collected using UAVs provides a great deal of detail through a method that reduces the amount of atmospheric scattering typically found in orbital satellite imagery. Temporal resolution of UAV imagery is much higher compared with other types of sensors and as such provides an advantage for examining surface features that change rapidly such as croplands. This in turn facilitates efficient monitoring of crops and pesticide management. This study employs UAV imagery collected from a quadcopter multirotor platform GEMS sensor with visible near-infrared cameras at .08 meter spatial resolution. The images were collected at an altitude of 60 meters between June and September 2015 at a temporal frequency of once a week. The study was conducted in a local community garden on the south side of the City of Eau Claire, Wisconsin. The 2,100 square meters garden is divided into approximately 30 plots with multiple crop types. Aerial imagery files were imported into PIX4D, along with a flight log, for mosaicking. Once the mosaic was completed, Ground Control Points (GCPs) were utilized to geometrically correct the images before further analysis. Prior to image segmentation, several physiological indices were generated at a high temporal frequency to gauge the growth rate of crops and assess their moisture, nutrient, and other climatic ingredients. Some of the indices included are Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), and Green Chlorophyll Index (CIgreen). Image objects were created in eCognition Developer using multiresolution segmentation algorithm. Image objects were then used to extract zonal statistics in order to compare the reflectance from the multispectral imagery with the vegetation values in the NDVI, GNDVI and CIgreen indices. R squared values for the relationship between vegetation indices and crop species health and ranged between .10 and .93. The precision agriculture-modeling framework developed in this study is highly invaluable to farmers, as it has been shown to increase crop growth information flows to farmers at very high temporal frequency.