ProTractor: A Lightweight Ground Imaging and Analysis System for Early-Season Field Phenotyping

Nico Higgs, Blanche Leyeza, Jordan Ubbens, Josh Kocur, William van der Kamp, Theron Cory, Christina Eynck, Sally Vail, Mark Eramian, Ian Stavness; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Abstract


Acquiring high-resolution images in the field for image-based crop phenotyping is typically performed by complicated, custom built "pheno-mobiles." In this paper, we demonstrate that large datasets of crop row images can be easily acquired with consumer cameras attached to a regular tractor. Localization and labeling of individual rows of plants are performed by a computer vision approach, rather than sophisticated real-time geo-location hardware on the tractor. We evaluate our approach for cropping rows of early-season plants from a Brassica carinata field trial where we achieve 100% recall and 99% precision. We also demonstrate a proof-of-concept plant counting method for our ProTractor system using an object detection network that achieves a mean average precision of 0.82 when detecting plants, and an R2 of 0.89 when counting plants. The ProTractor design and software are open source to advance the collection of large outdoor plant phenotyping datasets with inexpensive and easy to use acquisition systems.

Related Material


[pdf]
[bibtex]
@InProceedings{Higgs_2019_CVPR_Workshops,
author = {Higgs, Nico and Leyeza, Blanche and Ubbens, Jordan and Kocur, Josh and van der Kamp, William and Cory, Theron and Eynck, Christina and Vail, Sally and Eramian, Mark and Stavness, Ian},
title = {ProTractor: A Lightweight Ground Imaging and Analysis System for Early-Season Field Phenotyping},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}