Field Boundary Detection and Distance-Based Analysis for Bale Yield Prediction in Round Bale Harvesting

Agricultural machinery contractors, who often harvest multiple fields per day, require accurate yield predictions to estimate job times. Differences between predicted and actual yield can cause operators to miscalculate required inputs (e.g. plastic, net, fuel), potentially leading to costly idle periods and delays to subsequent field operations. This paper presents a coordinate-based method for predicting the number of bales remaining in a field, ahead of a tractor towing a round bale combination baler-wrapper implement, that forms cylindrical bales from swathes of mowed grass. The algorithm approximates the field area through a coverage mapping solution to detect a loop of the field boundary. The resultant area is used to estimate the total swathe distance, fitted from regression analysis on historical field areas (R2 = 0.986). The algorithm subsequently processes the observed bale distances to iteratively update predictions on the bales remaining in the field. The proposed method is validated on georeferenced time series tractor and baler data aggregated from a five-month-long measurement campaign of silage baling in Galway, Ireland. The mean absolute error, averaged across 120 fields, improves from an initial prediction error of 2.9 (±4.0) bales to 1.2 (±2.1) bales at 90% harvest completion. The main contribution of this research is a coordinate-based method for predicting the number of bales remaining in a field, based on evidence of a relationship between field yield and bale distances.