Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Desk-AId: Humanitarian Aid Desk Assessment with Geospatial AI for Predicting Landmine Areas

Flavio Cirillo, Gürkan Solmaz, Yi-Hsuan Peng, Christian Bizer, Martin Jebens

TL;DR

This work tackles the challenge of speeding up humanitarian mine action by improving desk assessments of landmine risk using geospatial AI. It introduces Desk-AId, a pipeline that fuses GIS-derived features, socio-economic data, and historical conflict information with machine learning, including a novel hard-negative sampling strategy and graph-based models. The approach is validated on Afghanistan data at country-wide scale and in uncharted study areas, showing strong performance with RF, FNN, and GCNN configurations and enabling risk visualization in QGIS. The results suggest practical impact for prioritizing demining operations, reducing costs and time, and guiding field surveys, with ongoing field trials in Cambodia and plans for broader deployment.

Abstract

The process of clearing areas, namely demining, starts by assessing and prioritizing potential hazardous areas (i.e., desk assessment) to go under thorough investigation of experts, who confirm the risk and proceed with the mines clearance operations. This paper presents Desk-AId that supports the desk assessment phase by estimating landmine risks using geospatial data and socioeconomic information. Desk-AId uses a Geospatial AI approach specialized to landmines. The approach includes mixed data sampling strategies and context-enrichment by historical conflicts and key multi-domain facilities (e.g., buildings, roads, health sites). The proposed system addresses the issue of having only ground-truth for confirmed hazardous areas by implementing a new hard-negative data sampling strategy, where negative points are sampled in the vicinity of hazardous areas. Experiments validate Desk-Aid in two domains for landmine risk assessment: 1) country-wide, and 2) uncharted study areas). The proposed approach increases the estimation accuracies up to 92%, for different classification models such as RandomForest (RF), Feedforward Neural Networks (FNN), and Graph Neural Networks (GNN).

Desk-AId: Humanitarian Aid Desk Assessment with Geospatial AI for Predicting Landmine Areas

TL;DR

This work tackles the challenge of speeding up humanitarian mine action by improving desk assessments of landmine risk using geospatial AI. It introduces Desk-AId, a pipeline that fuses GIS-derived features, socio-economic data, and historical conflict information with machine learning, including a novel hard-negative sampling strategy and graph-based models. The approach is validated on Afghanistan data at country-wide scale and in uncharted study areas, showing strong performance with RF, FNN, and GCNN configurations and enabling risk visualization in QGIS. The results suggest practical impact for prioritizing demining operations, reducing costs and time, and guiding field surveys, with ongoing field trials in Cambodia and plans for broader deployment.

Abstract

The process of clearing areas, namely demining, starts by assessing and prioritizing potential hazardous areas (i.e., desk assessment) to go under thorough investigation of experts, who confirm the risk and proceed with the mines clearance operations. This paper presents Desk-AId that supports the desk assessment phase by estimating landmine risks using geospatial data and socioeconomic information. Desk-AId uses a Geospatial AI approach specialized to landmines. The approach includes mixed data sampling strategies and context-enrichment by historical conflicts and key multi-domain facilities (e.g., buildings, roads, health sites). The proposed system addresses the issue of having only ground-truth for confirmed hazardous areas by implementing a new hard-negative data sampling strategy, where negative points are sampled in the vicinity of hazardous areas. Experiments validate Desk-Aid in two domains for landmine risk assessment: 1) country-wide, and 2) uncharted study areas). The proposed approach increases the estimation accuracies up to 92%, for different classification models such as RandomForest (RF), Feedforward Neural Networks (FNN), and Graph Neural Networks (GNN).
Paper Structure (19 sections, 22 figures, 2 tables)

This paper contains 19 sections, 22 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Geospatial AI
  4. Agriculture Mining
  5. Landmine Detection
  6. Desk-AId pipeline overview
  7. Building the dataset
  8. Source data to features
  9. Dataset for ML training
  10. Study Areas
  11. Machine learning models
  12. Experimental Result Analysis
  13. Evaluating the approaches for negative sampling
  14. Testing the Study Areas
  15. Adding features to the dataset
  16. ...and 4 more sections

Figures (22)

  • Figure 1: Break-down of demining operations into their major phases. Desk-AId falls into the phase of "Desk Assessment'.
  • Figure 2: Desk-AId pipeline overview from data sources till risk assessment
  • Figure 3: Landmine distribution and the two study areas in Afghanistan country land. Green area is the study area one (SA1) and orange one denotes study area two (SA2) that is around Kabul. Minefields are marked in red.
  • Figure 4: Hard negative samples generated around the positive samples (landmine polygons colored in red).
  • Figure 5: Box plot of numeric features of the two study areas SA1 and SA2.
  • ...and 17 more figures