Time series analysis of agricultural machines for automated field task detection based on a deep-learning approach

Documentation of field tasks for tractors is a manual, error prone task for farmers. We address this problem by exploiting the data measured by tractors and classifying the tractor’s work type with deep learning approaches: CNN and GRU. Compared to our and other previous approaches, the time-series based evaluation shows promising performance. At the same time, we show that the approach still lacks generalization potential concerning different farms and machine vendors. More data from different farms and machine types is still missing for sufficient training.