HeLoDL: Hedgerow Localization Based on Deep Learning

Accurate localization of hedges in 3D space is a key step automatic pruning. However, due to the irregularity hedge shape, accuracy based on traditional algorithms poor. In this paper, we propose deep learning approach bird’s-eye view overcoming problem, which call HeLoDL. Specifically, first project point cloud top-down as single image and, then, augment with morphological operations and rotation. Finally, trained convolutional neural network, HeLoDL, transfer learning, regress center axis radius hedge. addition, an evaluation metric OIoU that can respond error, well circle error integrated way. our test set, HeLoDL achieved 90.44% within tolerance, greatly exceeds 61.74% state-of-the-art algorithm. The average 92.65%; however, best conventional algorithm 83.69%. Extensive experiments demonstrated shows considerable spatial irregular models.