Description
In contrast to traditional data collection methods that require manual sampling, vegetative index (VI) maps derived from unmanned aerial vehicles (UAV) imagery are a potential tool to characterize temporal and spatial treatment effects in a more efficient and non-destructive way. Remotely-sensed reflectance data from a growing corn crop contains pixel values associated with the above-ground plant tissue (e.g., leaves, stalks, tassels) and the underlying soil features. Background soil reflectance data potentially reduces the effectiveness of VI for characterizing crop responses to experimental treatments. Removing background soil image pixels from the larger image dataset should improve that effectiveness. The objective of this study was compare the effectiveness of filtered and non-filtered VI maps in characterizing phenotypic responses of corn to fertilizer treatments. Three large scale field trials (12 to 20 ha) involving either sulfur or nitrogen fertilizer treatments were used for the study. Imagery was collected using a DJI Matrice 200 series UAV equipped with either a consumer RGB camera or a camera modified to capture NIR. Flights were conducted at corn growth stages V6, V10, and R4. The individual images were stitched into orthomosaic and image postprocessing was performed to calculate RGB (400-700 nm), and near-IR (700 to 1100 nm) based VIs. After performing image classification to separate plant from soil pixels, soil background was removed, and vegetative index values corresponding only to the plants were considered for the next steps. Analysis of variance and treatment contrasts were performed using filtered and non-filtered datasets. Furthermore, a regression analysis was performed to investigate the feasibility of VIs to estimate grain yield. Results suggest that removing soil background improves the characterization of corn responses to experimental treatments visually and statistically. R2 values between grain yield and VIs increased up to 0.4 after filtering soil background.
Start Date
11-2019
Document Type
Other
Keywords
Agronomy, UAV Based Remote Sensing, Vegetative Indices
Session List
Poster
Effects of Removing Background Soil Reflectance Pixels from Vegetative Index Maps for Characterization of Corn Responses to Experimental Treatments
In contrast to traditional data collection methods that require manual sampling, vegetative index (VI) maps derived from unmanned aerial vehicles (UAV) imagery are a potential tool to characterize temporal and spatial treatment effects in a more efficient and non-destructive way. Remotely-sensed reflectance data from a growing corn crop contains pixel values associated with the above-ground plant tissue (e.g., leaves, stalks, tassels) and the underlying soil features. Background soil reflectance data potentially reduces the effectiveness of VI for characterizing crop responses to experimental treatments. Removing background soil image pixels from the larger image dataset should improve that effectiveness. The objective of this study was compare the effectiveness of filtered and non-filtered VI maps in characterizing phenotypic responses of corn to fertilizer treatments. Three large scale field trials (12 to 20 ha) involving either sulfur or nitrogen fertilizer treatments were used for the study. Imagery was collected using a DJI Matrice 200 series UAV equipped with either a consumer RGB camera or a camera modified to capture NIR. Flights were conducted at corn growth stages V6, V10, and R4. The individual images were stitched into orthomosaic and image postprocessing was performed to calculate RGB (400-700 nm), and near-IR (700 to 1100 nm) based VIs. After performing image classification to separate plant from soil pixels, soil background was removed, and vegetative index values corresponding only to the plants were considered for the next steps. Analysis of variance and treatment contrasts were performed using filtered and non-filtered datasets. Furthermore, a regression analysis was performed to investigate the feasibility of VIs to estimate grain yield. Results suggest that removing soil background improves the characterization of corn responses to experimental treatments visually and statistically. R2 values between grain yield and VIs increased up to 0.4 after filtering soil background.