Location: Northern Great Plains Research Laboratory
Title: Corn early-season stand count and spatial distribution from UAV imagery using open-source ImageJAuthor
PATHAK, HARSH - NORTH DAKOTA STATE UNIVERSITY | |
CANNAYEN, IGATHINATHANE - NORTH DAKOTA STATE UNIVERSITY | |
FLORES, PAULO - NORTH DAKOTA STATE UNIVERSITY | |
SHAJAHAN, SUNOJ - CORNELL UNIVERSITY | |
Archer, David |
Submitted to: Meeting Abstract
Publication Type: Abstract Only Publication Acceptance Date: 8/29/2023 Publication Date: N/A Citation: N/A Interpretive Summary: Technical Abstract: Assessing plant stand count and spatial distribution early in the season is important as it helps farmers to make appropriate management decisions, and to evaluate planter efficiency and seed quality. Traditional stand count assessment methods are laborious, expensive, time-consuming, and prone to errors. Most commercially available farm management software imposes specific conditions on input imagery, processes data on the cloud leading to data security and privacy issues, and requires an annual subscription renewal. Information on the spatial distribution of plants is highly scarce. Therefore, an open-source image processing user-coded ImageJ plugin to count the number of plants and evaluate their spatial distribution of plants in the field was developed. The plugin requires color (RGB) input imagery collected from unmanned aerial vehicles (UAV), plant-to-plant and row-to-row spacing, number of regions of interest (ROI) per row, ROI width, and segmentation method. The excess green vegetation method was used to segment the crops and the profile-plot method to identify the crop rows. A novel “sliding and shifting ROI (SSROI)” method was developed that can efficiently shift to accommodate and count the plants along the rows even when plants deviate from a straight row. Using the centroid distances between the adjacent plants, a color-coded spatial distribution map visualizing the skips (singles, doubles, and multiple), doubles, and ideal spacings was developed. Textual outputs of stand count and plant spatial distribution (percent ideals, skips, doubles) along with overall map visualization would serve for site-specific management decisions. The plugin efficiently counted corn plants with an accuracy of > 97% and an average computational speed of 174 plants/s. The developed methodology can be extended to evaluate the stand count of other row crops such as sunflowers and safflowers. |