End-to-end deep learning approach to automated phenotyping of greenhouse-grown plant shoots

Description

e2e_predict is a demo software tool provided for demonstration of end-to-end approach to automated estimation of plant traits from RGB image of greenhouse-grown plants.

How to use?



The exectuable tool should be run from Windows command line using the following syntax:

e2e_predict "network_model_file_name" "image_file_name" "target_trait_name"

The example below demonstrate the application of the e2e_predict tool for computation of the plant "area" trait from the image "arab_top.png" stored in the subfolder "images" using the pretrained model "area_arab_top.mat" stored in the subfolder "models":

e2e_predict "models\area_arab_top.mat" "images\arab_top.png" "area"

Please, use e2e models that are complementary to the plant images and traits, e.g.,

area_arab_top.mat: for prediction of the "area" trait from arabidopsis top-view images, green_avg_barley_side.mat: for prediction of the "average green color" trait from barley side-view images, convex_hull_maize_top.mat: for prediction of the "convex hull area" trait from maize top-view images,

... and so on.

The list of traits:

area: area of the plant region in pixel^2,
convex_hull: area the convex hull of the plant in pixel^2,
height: height (i.e. vertical min/max dimension) of the plant region in pixels,
width: width (i.e. vertical min/max dimension) of the plant region in pixels,
h_99: 99% percentile of the vertical distribution of plant pixels (similar to height but more robust w.r.t. noise),
w_99: 99% percentile of the horizontal distribution of plant pixels (similar to wdith but more robust w.r.t. noise),
red_avg: average red color of the plant region,
green: average green color of the plant region,
blue: average blue color of the plant region.

The repository includes pretrained e2e models for the following five plant types, camera view (re. optical setups):

- arabidopsis top view images
- barley top view images
- barley side view images
- maize top view images
- maize side view images

Files of all pretrained models are stored in the "models" subfolder.
Files of example images are stored in the "images" subfolder.

Downloads

Download the e2e demo tool for Windows x64


By downloading, installing and/or using this software the user agrees with the following license conditions.

This software requires MATLAB run time libraries (2024a) to be installed:

MATLAB run time for Windows x64 (~2GB).

Windows users should also install the following run-time components if not yet installed: https://www.microsoft.com/en-us/download/details.aspx?id=48145.

Additional information on dependencies of the MATLAB run-time components in Linux can be found here: https://de.mathworks.com/matlabcentral/answers/358052-is-there-a-list-of-matlab-runtime-dependencies.

General hardware requirements for MATLAB applications can be found here: https://de.mathworks.com/support/requirements/matlab-system-requirements.html.
For processing of image files of the 4-8 MP size such as provided with the supplementary data, 8GB or more RAM is recommended.

Quick Start



References

[1] Gladilin et al. (2024) End-to-end deep learning approach to automated phenotyping of greenhouse-grown plant shoots.



© 2024, Image Analysis Group, IPK Gatersleben Impressum
Leibniz Institute of Plant Genetics and Crop Plant Research (IPK)
OT Gatersleben
Corrensstrasse 3
06466 Stadt Seeland,
Telefon: +49 (0)39482 50
Telefax: +49 (0)39482 5139
E-Mail: info@ipk-gatersleben.de
Internet: www.ipk-gatersleben.de