Recognition and Application Systems for Weed Control in the US.

Recognition Systems for Weed Control

Lie Tang, Iowa State University

Robotic technologies have started to revolutionize some agricultural field operations, for instance seeding and harvesting using GPS-guided auto-steer field equipment. However, agricultural field robotics by far has not reached its full potential, especially in utilizing robotic technology at the individual plant level, such as robotic weed removal. There are several primary limiting factors of this type of in-field selective and precise application, among which the development of high-throughput and robust weed/crop recognition systems is the most prominent. First, the operational context is semi-structured in which mixtures of undesired plant populations (i.e. weeds) exist along with semi-regularly spaced crop plants.  This situation demands sophisticated sensing systems and pattern recognition algorithms that meet real-time operational constraints. Second, the robotic weeding system works under environmental conditions (sunlight, wind, temperature, humidity) that are time-varying, imposing robustness challenges to its sensing system. The viability of a robotic weed control will depend on machine sensor system differentiation of crop and weed plants, localization of target plants, and precise simultaneous control of mobile robotic platform posture and end-effector motion with a chief performance goal to terminate weeds while not damaging valuable crop plants. This presentation will be focusing on some of the most recent development of sensing technologies for weed/crop sensing, particularly in the machine vision related areas.