nasa-nccs-hpda/senegal-lcluc-tensorflow

Senegal LCLUC TensorFlow # Senegal LCLUC TensorFlow Python library to process and classify remote sensing imagery by means of GPUs and CPU parallelization for high performance and commodity base environments. This repository focuses in using CNNs for the inference of very high-resolution remote sensing imagery in Senegal. Figure 1. Study area of this repository. We are currently working on tutorials and documentations. Feel to follow this repository for documentation updates and upcoming tutorials. ## Science Questions - Can 2m VHR imagery enhance our understanding of changes in the extent, intensity and land use of agriculture and forestry in Senegal? - How can we better take advantage of NASA’s HEC resources to apply Deep learning for land cover change monitoring? - Can we scale-up Unet CNNs to compensate for the diversity of landscapes and images to map regional land cover? Figure 2. Wet and dry seasonality for cluster of typical fields in Senegal, Photos from Collaborator Gray Tappan ## Getting Started 1) Log on to the NASA ADAPT server ```bash ssh user@adaptlogin.nasa.gov ``` 2) Connect to PRISM GPU Cluster ```bash ssh gpulogin1 ``` 3) Request computational resources with Slurm (see NCCS guide to Slurm on ADAPT). For example, to request a quick interactive Slurm job: ```bash salloc -G1 -J composite-text -c 10 ``` which requests an interactive job (salloc), with 1 GPU (-G), a job name (-J) of "composite-text", and 10 CPU cores (-c). 4) Create personal directories to work within: - Confirm that the $USER environment variable is set to your username ```bash echo $USER ``` which should return your username. * Create a "development" directory ```bash mkdir -p /explore/nobackup/people/$USER/development ``` 5) Clone the GitHub ```bash cd /explore/nobackup/people/$USER/development # Navigate to development folder git clone github.com ``` 6) Load necessary module(s). ```bash module load singularity ``` ### Bonus Tips - If y …