TCMFE-Net

For Tree species Classification.
GMFE-Net (PyTorch) classification version.
The previous GMFE-Net was implemented based on Keras 2.6 (https://github.com/LiTao-0797/GMFE-Net).

Installation

OS: Windwos 10 or 11
RAM: 64GB or above
GPU: NVIDIA Quadro RTX6000 or similar
IDE: Spyder (in Anaconda)
Environment: Python 3.9.21, torch 2.7.1+cu118, torch_cluster 1.6.3+pt27cu118, torch-geometric 2.6.1, numpy 1.26.3, laspy 2.5.4, pandas 2.2.3, scikit-learn 1.6.1, seaborn 0.13.2

Usage

TocknerPLS dataset

1. First, download the dataset from https://zenodo.org/records/10035928, then extract "_species_trees.zip" into a newly created "species_trees" folder, and finally place "species.txt" into this folder.
   
2. Open the file "step1_data_preparation.py" in Spyder, then click "Run" to start the data preprocessing. Please note to modify the dataset storage path.
   
3. After completing the data preprocessing, open the file "step2_train_test_species.py" in Spyder, then click "Run" to begin training and testing on the TocknerPLS dataset. Please note the configuration of parameters such as "USE_FEATURES" and "sampling_method".
   

Thanks to Tockner et al. for providing the dataset (https://doi.org/10.1016/j.jag.2025.104502).

STPCTLS dataset

1. First, download the dataset from https://doi.org/10.25625/FOHUJM (Seidel et al.) or https://github.com/said-ohamouddou/STFT-KAN-liteDGCNN (Ohamouddou et al.). The STPCTLS dataset we organized is also available here: https://doi.org/10.5281/zenodo.18626586.
   
2. Open the file "step1_data_preparation.py" in Spyder, then click "Run" to start the data preprocessing. Please note to modify the dataset storage path.
   
3. After completing the data preprocessing, open the file "step2_train_test_STPCTLS.py" in Spyder, then click "Run" to begin training and testing on the STPCTLS dataset. Please note the configuration of parameters such as "EXPERIMENT_MODE" and "sampling_method".
   

Thanks to Seidel et al. for providing the dataset (https://doi.org/10.3389/fpls.2021.635440).
Thanks to Ohamouddou et al. for organizing and naming the dataset (https://doi.org/10.1016/j.patcog.2025.112584).

Weights download URL

Weights download URL: https://drive.google.com/drive/folders/1OaEy9zekFIVGHujiXYgGcGauRZK4cuiq?usp=sharing

Acknowledgments

This project wouldn't have been possible without the support and contributions of several individuals and resources.
Thanks to (in no particular order):

• https://github.com/charlesq34/pointnet
• https://github.com/charlesq34/pointnet2
• https://github.com/HuguesTHOMAS/KPConv
• https://github.com/hkust-vgd/shellnet
• https://github.com/QingyongHu/RandLA-Net
• https://github.com/leofansq/SCF-Net
• https://keras.io/examples/vision/pointnet/