Deep Conditional Census-Constrained Clustering (DeepC4) for Large-scale Multi-task Spatial Disaggregation of Urban Morphology

This Zenodo record contains the datasets of our research (Spatial Disaggregation of Rwandan Building Exposure and Vulnerability via Weakly Supervised Conditional Census-Constrained Clustering (C4) using Earth Observation Data) submiited for American Geophysical Union Annual Meeting 2024 to be held in Washington, D.C. on 9th-13th of December 2024. The GitHub repository of MATLAB & Python codes can be accessed here: github.com/riskaudit/DeepC4. If you have any inquiries or would like to access any related materials, please feel free to visit my website (joshuadimasaka.com) or our project website (riskaudit.github.io), follow our project's GitHub repository (github.com/riskaudit), or send an email to jtd33@cam.ac.uk.

History of Versions: 

• 1.0.0 - Initial upload (C4)
• 1.1.0 - Updated output_yMapsAndQGISStyles.zip (DeepC4)
  • Investigated several dimensional reduction algorithms.
  • Incorporated AutoEncoders (deep, flexible, and nonlinear capability) in achieving an efficient latent representation that is derived from multiple EO signals and informative for clustering of target urban morphology clustering classes.
• 1.2.0 - Updated output_yMapsAndQGISStyles.zip (DeepC4) and Added output_PrecisionMapsAndQGISStyles.zip.
  • Significantly improved the constrained clustering algorithm (i.e., Minimum Cost Flow) by using the side information as initial point for clustering iteration.
  • Improved metrics with consideration of imbalance class labels.
  • Refined training set (to 20 sectors) that provides limited and weak supervision.
  • Revised mapping scheme between target labels and side information.
• 2.0.0 - Corrected output_yMapsAndQGISStyles.zip (DeepC4)
  • Made a correction related to the number of dwellings considered.
  • Uploaded additional files (e.g., input preprocessed data, resulting trained model learning vs epoch, and other figures and flowcharts used for manuscript write-up).
  • Added AGU official poster.