Published March 8, 2021 | Version 2.14.00
Software Open

Leaf Vein Network CNN Analysis Software v2.14

Creators

  • 1. University of Oxford
  • 2. University of California, Berkeley

Description

Software to extract and analyse leaf vein networks using convolutional neural networks.

  • LeafVeinCNN.mlappinstall - installs version 2.14 of the software as a matlab app. The package includes the original set of trained CNN models (at 1.68 µm resolution). Requires Matlab 2020b or later, and the Deep Learning Toolbox, Image Processing Toolbox, and Parallel Computing Toolbox for full functionality. In addition, the support package 'Deep Learning Toolbox Converter for TensorFlow Models' needs to be installed using the Add-On Explorer
  • LeafVeinCNN.exe - installs version 2.14 of the software and trained CNN models as a standalone package for Windows 11. This will automatically download the Matlab runtime library from the web during installation. This update includes a bug fix that prevented the software from running without manually adding the Reference Table containing all the calibration units.
  • LeafVeinCNN_Manual v2.pdf - A user manual describing installation and use of the software.

This update includes additional options to help segment vein networks from monocot leaves or leaves with branching-only networks by extrapolating from the peripheral free-ending veins to the lamina margin. This is currently a Beta version that works on a small subset of images we have tested.  

  1. To run the normal Hierarchical Loop Decomposition on loopy leaves, use the Prune boundary and GCC options.
  2. For monocots or predominantly tree like leaves use the Add Boundary option and turn off the other two.

The new 'script' checkboxes are also Beta versions that are exploring faster code implementations including parallel options.

Feedback welcome!

Bug fixes:

Binary images for masks, veins, ground truths etc should now load properly (thanks to Jingchao Zhao for flagging this)

Notes

Additional funding from: Human Frontier Science Program (RGP0053/2012), Leverhulme Trust (RPG-2015-437), NSF: DEB-2025282 and RoL:FELS:RAISE DEB-1840209

Files

LeafVein_Manual v2.pdf

Files (2.3 GB)

Name Size Download all
md5:c4d71c132f325c873f3edd372bc8683e
76.5 MB Preview Download
md5:ef2b884cc9a586f177806abbaeca6845
1.1 GB Download
md5:71a66b8c595fe504dfab329bcb0506f5
1.1 GB Download

Additional details

Funding

Towards a more predictive community ecology: integrating functional traits and disequilibrium NE/M019160/1
UK Research and Innovation
BIODIVERSITY AND LAND-USE IMPACTS ON TROPICAL ECOSYSTEM FUNCTION (BALI) NE/K016253/1
UK Research and Innovation
GEM-TRAIT – GEM-TRAIT: The Global Ecosystems Monitoring and Trait Study: a novel approach to quantifying the role of biodiversity in the functioning and future of tropical forests. 321121
European Commission

References

  • Xu, H., Blonder, B., Jodra, M., Malhi, Y. and Fricker, M.D. (2020) Automated and accurate segmentation of leaf venation networks via deep learning. New Phytologist 229, 631-648. Doi: 10.1111/nph.16923.
  • Blonder, B., S. Both, M. Jodra, H. Xu, M. Fricker, I. S. Matos, N. Majalap, D. F. R. P. Burslem, Y. Teh and Y. Malhi (2020) Linking functional traits to multiscale statistics of leaf venation networks. New Phytol. 228, 1796-1810. Doi: 10.1111/nph.16830.