🌳 Cerrado Tree Species Identificator

Welcome to the Cerrado Tree Species Identificator project!

This repository introduces a novel image preprocessing pipeline that fuses multispectral images, tailored for native species of the Cerrado biome.
Using our local dataset, the models provided in models can detect over 18 different species. Feel free to use them as you like!

Our pipeline delivers a 4.2% improvement in detection accuracy compared to state-of-the-art RGB-only approaches.

Note: Running the preprocessing notebooks requires a machine with a GPU for efficient processing.

---

🚀 Getting Started

Follow the steps below to set up and run the project:

1. Create a virtual environment

We recommend using venv or conda to manage dependencies.

python -m venv your_env
source your_env/bin/activate

2. Install dependencies

• Install exiftool for extracting image metadata:

sudo apt update && sudo apt upgrade 
sudo apt install libimage-exiftool-perl

• Then, install Python packages:

pip install -r requirements.txt

---

⚙️ CUDA Setup (GPU Acceleration)

To run preprocessing efficiently, you’ll need a working CUDA environment. Here’s how to set it up:

1. Check for a compatible NVIDIA GPU

nvidia-smi

If your GPU appears in the output, you're good to go!

2. Install NVIDIA Drivers

Download and install the latest drivers for your GPU from: https://www.nvidia.com/Download/index.aspx

3. Install CUDA Toolkit and cuDNN

Choose a version compatible with your PyTorch installation. Example for CUDA 11.8:

# Ubuntu example
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
sudo apt update
sudo apt install -y cuda

After installation, add CUDA to your path:

echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
source ~/.bashrc

4. Verify CUDA is working in Python

import torch
print(torch.cuda.is_available())  # Should return True

---

🧪 Running the Pipeline

After a functional GPU and an appropriate environment are set up, you're ready to run the project!

Step 1: Preprocess images

Open and run the python script:

./src/preprocessing/main.py

Preprocessed images will be saved to:

{path-to-dataset}/{specified-date}

Step 2: Fuse multispectral channels

Open and run the python script:

./src/image_fusion/main.py

Fused images will be saved to:

{path-to-dataset}/{specified-date}/fused-imgs

Step 3: Generate the vegetation indexes and the combined images

Open and run the python script:

./src/vegetation_index/main.py

The vegetation indexes images will be saved to:

{path-to-dataset}/{specified-date}/ndre
{path-to-dataset}/{specified-date}/ndvi

The vegetation indexes images combined/stacked with the RGB and Fused images will be saved to:

{path-to-dataset}/{specified-date}/fused-ndre
{path-to-dataset}/{specified-date}/fused-ndvi

{path-to-dataset}/{specified-date}/rgb-ndre
{path-to-dataset}/{specified-date}/rgb-ndvi

Step 4: Train the models

The models are trained in batch for all the image types and variation availabe. For train a specific model, go to src/models/{selected-model}/train and execute:

train.sh

Artifacts generated during trainig such as images, weights, charts and etc. will be saved in runs/{model-name}/no-augmentation/{image-type}

Step 5: Evaluate training

To evaluate the executed trainig in the test set, go to src/models/{selected-model}/eval and execute:

eval.sh

The artifacts generated such as images and charts will be saved in reports/{model-name}/test_no_augmented_{image-type}

---

That’s it! You're ready to explore tree species detection in the Cerrado using fused multispectral data.