The Queensland University of Technology (QUT) has investigated the application of machine vision and learning techniques for the estimation of on farm pasture biomass and the automated identification of features in agricultural landscapes such as trees, houses and roads.
Sensor data from the WorldView-2 satellite provided high resolution imagery (0.46 meter pixels) in ten spectral bands for the calculation of vegetation indices, see Figure 1. Particular focus was placed upon the Normalized Difference Vegetation Index (NDVI) that uses visible and near-infrared spectra to give an indication of the condition of vegetation from satellite imagery.
This data can be used to investigate pasture performance and grazing capacity. Time series data was used to develop a model that is capable of predicting change in pasture biomass and condition; the model has the capacity to fine tune itself (learn) over time.
The efficacy of this model has been increased through the elimination of non-pasture elements of the agricultural landscape. Further development is required to make these biomass estimates quantitative, integrate them into the data dashboard and create an automated workflow for data streams purchased from imagery outlets. Other technology systems, such as unmanned autonomous vehicles or ground vehicles, may provide better or complementary pasture monitoring solutions in some situations and are becoming more affordable and accessible.