This online software is used to reconstruct damaged leaves based on pretrained deep-learning models and quantify the degree of leaf damage through image processing techniques. Navigate to 'About' tab for the user guide.
For the detailed method, see: Wang, Z., Jiang, Y., Diallo, A. B., & Kembel, S. W. (2024). Deep learning- and image processing-based methods for automatic estimation of leaf herbivore damage. Methods in Ecology and Evolution, 15, 732–743.
'HerbiEstim' is an online software designed for automatic estimation of leaf herbivore damage. The software takes advantage of pretrained deep learning models to reconstruct herbivore damaged leaves and applied image processing techniques to quantify the percentage of leaf damage. Due to the limited computing resources, this online software only takes <50 images per run. For a large image volume, please use the offline version (https://github.com/ZihuiWang1/HerbiEstim)
The input should be scanned or photographed leaf images with blank background. Both single-leaf or multiple-leaf images are acceptable. However, in the case of multiple-leaf images, overlaps and contacts among leaves are prohibited. Download 'Example images' below to understand the image requirements.
Example imagesTwo types of pretrained models are available. The Universal model was trained using a variety of leaf shapes and thus can be used to reconstruct plant leaves for different species. Case-specific models is useful for studies focusing on single plant species or unusual leaf shapes.
The resolution of scanned images is required for estimating exact leaf area loss. If resolution information is not available, only the percentage of leaf area loss is estimated.
Two types of output are generated for download.
Output Images: a combined image of the reconstructed (on the right) and raw leaves (on the left).
Output Table: The estimated area and percentage of leaf damage. Columns are tag of images (img.tag), tag of individual leaves (ind.leaf), leaf area of damaged leaves (LA) and reconstructed intact leaves (intact.LA), and the percentage of leaf area loss (damage).
The software was programmed by Zihui Wang (wang.zihui@courrier.uqam.ca)
The method was developed by Zihui Wang, Yuan Jiang, Abdoulaye Baniré Diallo and Steven W. Kembel.
If you use this software, please cite:
Wang, Z., Jiang, Y., Diallo, A. B., & Kembel, S. W. (2024). Deep learning- and image processing-based methods for automatic estimation of leaf herbivore damage. Methods in Ecology and Evolution, 15, 732–743. https://doi.org/10.1111/2041-210X.14293
Wang, Z. (2024). HerbiEstim: A deep learning- and image processing-based software for automatic estimation of leaf herbivore damage (v1.0.0). Zenodo. https://doi.org/10.5281/zenodo.10514389