Detecting wires in the canopy of grapevines using neural networks: A robust and heuristic free approach

The location of wires in the canopy of grapevines is critical information required when planning the path of a robotic arm used in an automated vine pruning system. The cost of colliding with or inadvertently cutting a wire is severe and thus it is important to have a robust wire detection system that is capable of accurately locating wires. This paper proposes a system for detecting pixels with a high probability of being a wire in two dimensional Bayer images using neural networks. We are able to determine if a pixel belongs to a wire with 94% precision and can classify a pixel in 0.022 milliseconds.