Precision farming robots, which target to reduce the amount of herbicidesthat need to be brought out in the fields, must have the ability to identifycrops and weeds in real time to trigger weeding actions. In this paper, weaddress the problem of CNN-based semantic segmentation of crop fieldsseparating sugar beet plants, weeds, and background solely based on RGB data.We propose a CNN that exploits existing vegetation indexes and provides aclassification in real time. Furthermore, it can be effectively re-trained toso far unseen fields with a comparably small amount of training data. Weimplemented and thoroughly evaluated our system on a real agricultural robotoperating in different fields in Germany and Switzerland. The results show thatour system generalizes well, can operate at around 20Hz, and is suitable foronline operation in the fields.
展开▼
