Wind Farm Spotter is an inference engine for classifying the capacity of existing land-based U.S. Wind Farms and the potential capacity of unoccupied locations for new Wind Farm projects based on aerial satellite images.
- Google Earth Engine
- Google Drive
- Google Colab/JupyterNotebook
- Basemap
- ArcGIS API Service
- PyTorch 1.1 / Torchvision v0.3.0
- pytorchcv
- Fastai
- XQuartz (X11)
- Ubuntu 18.04.3
- Jetpack 4
- Flask, Python 3.6, pip3
- Kitura, Swift 5.0.1, Swift PM, swiftlang
- Virtualenv
- Docker Community Edition, Edge
Medium Post: https://medium.com/experimenting-with-deep-learning/spotting-potential-classifying-prime-areas-for-renewable-wind-energy-farms-with-computer-vision-3085018c821c (Draft)
Link: Here
Link: Here
Create a virtualenv environment:
virtualenv -p python3 <env name, e.g., wfs_py_env>
Activate the environment:
source <env name, e.g., wfs_py_env> && cd ~/<env name, e.g., wfs_py_env>**
source bin/activate
Update, Upgrade, and Purge Apt Dependencies:
sudo apt-get -y update
sudo apt-get -y upgrade
sudo apt-get -y purge \
python3-numpy python3-scipy \
python3-matplotlib python3-pandas python3-pil
Install Necessary Dependencies:
sudo apt-get -y install \
python3-pip build-essential python-dev git \
cython python-requests python-typing \
g++ cmake python-dev \
nodejs libssl1.0-dev nodejs-dev node-gyp npm freetype2-demos \
zlib1g-dev zip libjpeg8-dev libhdf5-dev \
libhdf5-serial-dev hdf5-tools \
pkg-config libfreetype6-dev libpng-dev \
libblas3 liblapack3 liblapack-dev libblas-dev \
gfortran htop
Install Spacy:
sudo pip3 install spacy==2.0.18
Download and Install PyTorch wheel for Jetson Devices + Torchvision v0.3.0 source and Install:
wget https://nvidia.box.com/shared/static/mmu3xb3sp4o8qg9tji90kkxl1eijjfc6.whl -O torch-1.1.0-cp36-cp36m-linux_aarch64.whl
sudo pip3 install numpy torch-1.1.0a0+b457266-cp36-cp36m-linux_aarch64.whl
sudo apt install libjpeg-dev libfreetype6-dev pkg-config libpng-dev
git clone -b v0.3.0 https://github.com/pytorch/vision torchvision
cd torchvision
sudo python3 setup.py install
cd ../
(Note: if you'd like to check that everything's woprking properly, test the install following instructions I.ve written over at Medium: https://medium.com/hackers-terminal/installing-pytorch-torchvision-on-nvidias-jetson-tx2-81591d03ce32)
Install pip requirements:
sudo pip3 install -r requirements.txt
Clone this repo to the virtualenv:
git clone -b wfs-py-edge https://github.com/codeamt/WindFarmSpotter app && cd app
And from here, you're ready to develop and test endpoints
Steps:
- Create an SSH Connection to the Jetson from Host
ssh -X -Y <user>:<ip_address>
- Open a seperate terminal to post an inference request
- From SSH terminal, cd into project directory, activate env, cd into app directory
- Run the engine server
- Make a GET request to index for more instructions and paths to test images
gnome terminal
cd ~/<env name, e.g., wfs_py_env> && source bin/activate && cd app
sudo python3 run.py
curl -X GET http:https://127.0.0.1:5000/
Sample POST request to predict endpoint:
curl -X POST http:https://127.0.0.1:5000/predict -H "Content-Type: application/json" -d '{"file":"engine/static/test_imgs/test_no_turbines_high_potential_5.jpg"}'
Steps:
- Create an port forwarded SSH Connection to the Jetson from Host:
- cd into project directory, activate env, cd into app directory:
- Run the engine server
ssh -X -Y -L 5555:localhost:5555 <user>:<ip_address>
cd ~/<env name, e.g., wfs_py_env> && source bin/activate && cd app
sudo python3 run.py
Visit localhost:5555 in your browser on the host machine, browse images, and click on one to get a prediction!
