spgriffin / ACM-DEV

code for the ACMDEV publication

LANDSAT-landstats (ACM-DEV paper edition)

We can use the files in this repository to predict population from satellite images.

LANDSAT lanstats is a supervised learning model to predict socio-economic characteristics from satellite data. We can easily modify this code to predict other socio-economic characteristics or use other satellite images. In addition to the convnet, I've provide files to construct the data.

The files of interest are:

• cnn.py: trains the convolutional neural network
• cnn_estimation.py: uses trained model for estimation
• data_cleaning.py: a class to merges satellite images with population databases (shapefiles)
• do_data_cleaning.py: runs data_cleaning.py to create training dataset
• validation.py: class to create validation dataset
• do_validation.py: creates validation data set
• do_postestimation.py: converts satellite images to predictions
• estimates.ipynb: generating estimates from predictions

In addition to the usual, the analysis relies on these packages

• Keras
• Google Earth Engine
• Shapely
• GeoPandas
• Rtree
• GDAL
• pyproj
• parmap
• h5py