TY - JOUR
T1 - Length phenotyping with interest point detection
AU - Vit, Adar
AU - Shani, Guy
AU - Bar-Hillel, Aharon
N1 - Funding Information: This research is supported by the Israel Innovation Authority through the Phenomics MAGNET Consortium, and by the ISF fund, under Grant No. 1210/18. We thank Ortal Bakhshian from Rahan Meristem and Lena Karol from Hazera Genetics for many helpful discussions and for providing the images and annotations. Publisher Copyright: © 2020 Elsevier B.V.
PY - 2020/9/1
Y1 - 2020/9/1
N2 - Plant phenotyping is the task of measuring plant attributes mainly for agricultural purposes. We term length phenotyping the task of measuring the length of a plant part of interest. The recent rise of low cost RGB-D sensors and accurate deep artificial neural networks provides new opportunities for length phenotyping. We present a general technique for length phenotyping based on three stages: object detection, point of interest identification, and a 3D measurement phase. We address object detection and interest point identification by training network models for each task, and develop a robust de-projection procedure for the 3D measurement stage. We apply our method to three real world tasks: measuring the height of a banana tree, the length and width of banana leaves in potted plants, and the length of cucumbers fruits in field conditions. The three tasks were solved using the same pipeline with minor adaptations, indicating the method's general potential. The method is stagewise analyzed and shown to be preferable to alternative algorithms, obtaining error of less than 10% deviation in all tasks. For leaves’ length and width, the measurements are shown to be useful for further phenotyping of plant treatment and mutant classification.
AB - Plant phenotyping is the task of measuring plant attributes mainly for agricultural purposes. We term length phenotyping the task of measuring the length of a plant part of interest. The recent rise of low cost RGB-D sensors and accurate deep artificial neural networks provides new opportunities for length phenotyping. We present a general technique for length phenotyping based on three stages: object detection, point of interest identification, and a 3D measurement phase. We address object detection and interest point identification by training network models for each task, and develop a robust de-projection procedure for the 3D measurement stage. We apply our method to three real world tasks: measuring the height of a banana tree, the length and width of banana leaves in potted plants, and the length of cucumbers fruits in field conditions. The three tasks were solved using the same pipeline with minor adaptations, indicating the method's general potential. The method is stagewise analyzed and shown to be preferable to alternative algorithms, obtaining error of less than 10% deviation in all tasks. For leaves’ length and width, the measurements are shown to be useful for further phenotyping of plant treatment and mutant classification.
KW - Interest points detection
KW - Length estimation
KW - Plant phenotyping
KW - RGB-D sensor
UR - http://www.scopus.com/inward/record.url?scp=85088132242&partnerID=8YFLogxK
U2 - https://doi.org/10.1016/j.compag.2020.105629
DO - https://doi.org/10.1016/j.compag.2020.105629
M3 - Article
SN - 0168-1699
VL - 176
JO - Computers and Electronics in Agriculture
JF - Computers and Electronics in Agriculture
M1 - 105629
ER -