Skip to content
Code for constructing the urban environments dataset and for land use classification via convolutional neural networks
Jupyter Notebook Python
Branch: master
Clone or download

README.md

Comparing urban environments using satellite imagery and convolutional neural networks

This repository contains code related to the paper Using convolutional networks and satellite imagery to identify patterns in urban environments at a large scale. A slightly modified version of the paper appears in the proceedings of the ACM KDD 2017 conference.

This repository contains the Python implementation of the data processing, model training, and analysis presented in the paper:

  • code to construct training and evaluation datasets for land use classification of urban environments is in the dataset-collection folder
  • Keras implementations of the convolutional neural networks classifiers used for this paper are in the classifier folder, along with Keras utilities for data ingestion and multi-GPU training (with a TensorFlow backend)
  • code to train and validate the models, and to produce the analysis and figures in the paper is in the notebooks in the land-use-classification folder.

After a convolutional network classifier is trained on satellite data in a supervised way, it can be used to compare city blocks (urban environments) across many cities by studying the features extracted for each satellite image.

Citation

If you use the code, data, or analysis results in this paper, we kindly ask that you cite the paper above as:

Using convolutional networks and satellite imagery to identify patterns in urban environments at a large scale. A. Toni Albert, J. Kaur, M.C. Gonzalez, 2017. In Proceedings of the ACM SigKDD 2017 Conference, Halifax, Nova Scotia, Canada.

You can’t perform that action at this time.