agroforest-forest mapping in Peru

code for agroforestry mapping in peru

How to use the code

0 set the conda environment

conda env create -f environment_for_agroforest.yml

1 prepare the data

1.1 Download the samples within the link (https://zenodo.org/records/13946752)

1.2 Download the Nicfi tiles that is overlay with the samples (https://www.planet.com/nicfi/)

we used 3m resolution Planerscope,but you are welcomed to apply 5m NICFI data to give a try!

you can download the data through Google Earth Engine

1.3 Stack the Nicfi data with samples to make the data ready for training

we saved the training dataset with .npy

2 training

2.1 change the dataroot with your dataset

2.2 ready for training

python train.py --model unet --alpha 0.1 --batch-size 16 --gpu-ids 0 --imsize 128 --lr 8e-7 --lr-policy poly --net resnet50 --num-threads 32 --niter 1000 --val-freq 2000 --print-freq 10 --checkpoints-dir your_pathway_for_model