Prediction of Soybean Plant Density Using a Machine Learning Model and Vegetation Indices Extracted from RGB Images Taken with a UAV

Soybean plant density is an important factor of successful agricultural production. Due to the high number of plants per unit area, early plant overlapping and eventual plant loss, the estimation of soybean plant density in the later stages of development should enable the determination of the final...

Full description

Saved in:
Bibliographic Details
Published in:Agronomy (Basel) 2020-08, Vol.10 (8), p.1108
Main Authors: Ranđelović, Predrag, Đorđević, Vuk, Milić, Stanko, Balešević-Tubić, Svetlana, Petrović, Kristina, Miladinović, Jegor, Đukić, Vojin
Format: Article
Language:English
Subjects:
Agricultural production
Cameras
Channels
Color imagery
Corn
Correlation coefficient
Correlation coefficients
Crops
Data collection
Developmental stages
Digital imaging
Global positioning systems
GPS
Learning algorithms
Machine learning
Open source software
Planting density
Public domain
RGB images
Root-mean-square errors
soybean
Soybeans
UAV
Unmanned aerial vehicles
Vegetation
vegetation indices
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Soybean plant density is an important factor of successful agricultural production. Due to the high number of plants per unit area, early plant overlapping and eventual plant loss, the estimation of soybean plant density in the later stages of development should enable the determination of the final plant number and reflect the state of the harvest. In order to assess soybean plant density in a digital, nondestructive, and less intense way, analysis was performed on RGB images (containing three channels: RED, GREEN, and BLUE) taken with a UAV (Unmanned Aerial Vehicle) on 66 experimental plots in 2018, and 200 experimental plots in 2019. Mean values of the R, G, and B channels were extracted for each plot, then vegetation indices (VIs) were calculated and used as predictors for the machine learning model (MLM). The model was calibrated in 2018 and validated in 2019. For validation purposes, the predicted values for the 200 experimental plots were compared with the real number of plants per unit area (m2). Model validation resulted in the correlation coefficient—R = 0.87, mean absolute error (MAE) = 6.24, and root mean square error (RMSE) = 7.47. The results of the research indicate the possibility of using the MLM, based on simple values of VIs, for the prediction of plant density in agriculture without using human labor.
ISSN:2073-4395
2073-4395
DOI:10.3390/agronomy10081108