Research Article: Extraction of Small Spatial Plots from Geo-Registered UAS Imagery of Crop Fields

Abstract

Experimental crop fields often contain large numbers of small spatial plots that require separate analysis. It can be difficult to precisely locate these plots in high-resolution imagery of the fields gathered by Unmanned Aircraft Systems (UAS). This prevents UAS imagery from being applied in High-Throughput Precision Phenotyping and other areas of agricultural research. If the imagery is accurately geo-registered, then it may be possible to extract plots from the imagery based on their map coordinates. To test this approach, a UAS was used to acquire visual imagery of 5 ha of soybean fields containing 6.0 m2 plots. Sixteen artificial targets were set up in the fields before flights to be used as Ground Control Points (GCPs) for geo-registration. Geo-registration accuracy was quantified based on the horizontal Root Mean Squared Error (RMSE) of targets used as checkpoints. Twenty test plots were extracted from the geo-registered imagery. Plot extraction accuracy was quantified based on the percentage of the desired plot area that was extracted. It was found that using four GCPs minimized the horizontal RMSE and enabled a plot extraction accuracy of at least 70%, with a mean plot extraction accuracy of 92%. Future work will focus on enhancing the plot extraction accuracy through additional image processing techniques so that it becomes sufficiently accurate for all practical purposes in agricultural research and potentially other areas of research.