Real-time Semantic Segmentation of Crop and Weed for Precision\n Agriculture Robots Leveraging Background Knowledge in CNNs

Abstract

Precision farming robots, which target to reduce the amount of herbicides\nthat need to be brought out in the fields, must have the ability to identify\ncrops and weeds in real time to trigger weeding actions. In this paper, we\naddress the problem of CNN-based semantic segmentation of crop fields\nseparating sugar beet plants, weeds, and background solely based on RGB data.\nWe propose a CNN that exploits existing vegetation indexes and provides a\nclassification in real time. Furthermore, it can be effectively re-trained to\nso far unseen fields with a comparably small amount of training data. We\nimplemented and thoroughly evaluated our system on a real agricultural robot\noperating in different fields in Germany and Switzerland. The results show that\nour system generalizes well, can operate at around 20Hz, and is suitable for\nonline operation in the fields.\n