ABSTRACT
This paper presents the evolution of a vision system dedicated to automatic weeding, initially implemented on a NVIDIA Jetson Xavier board. This evolution aims to take advantage of a new computing board able to implement efficient artificial intelligence oriented computations, keeping a low power consumption, and a low cost, developed in the ANDANTE project. The paper presents the automatic weeding tool, the existing vision system and the weeding data used to train the system. It also describes the specifications of the new board and the adaptation needed in order to integrate the previous algorithm in this new board. The results obtained during the first step of this integration are presented and compared to those obtained with the previous vision system. These new results are encouraging and rich in lessons for the future.
