Combine Harvester Unloading Event Inference Using GPS Data

Abstract. For typical wheat harvesting scenarios in modern agriculture, unloading events, either initiated by the combine harvester or the grain cart, serve as an important gateway to acquire insights on field efficiency and machine power efficiency, if the events are properly identified. When available, they also enable the possibility for product tracing and calibrations of yield monitors. Ideally, researchers could collect processed data from the yield monitor or raw data from the ISOBUS diagnostic port to extract auger status or auger flow rate for unloading events identification. However, interpreting messages from these sources are extremely challenging due to the proprietary nature of these messages. In this work, we took advantage of Global Position System (GPS) data instead and focused on the automatic identification of in-field unloading events from combine harvesters to tractor-hauled grain carts. We proposed a workflow that employed Interacting Multiple Model filtering and a rule-based algorithm to extract unloading events from GPS tracks. To verify the performance, we first applied the workflow to the GPS data collected from two combines and one tractor during 16 wheat harvest sessions to identify combine unloading events. Using the identified events, we then estimated the cumulative grain amount harvested during these harvesting sessions. The average absolute error of the estimated cumulative grain weights was within 10 percent of the actual cumulative grain amount recorded by weight ticket readings.