Effective Machine Learning Techniques for Runway Detection in Airfields

Strategic Insights -

A recent study published in Earth Science Informatics has investigated the use of machine learning (ML) algorithms for identifying runways and associated airfield features using open-source satellite imagery. The research, conducted by the Birla Institute of Technology, employed three data pathways, including SAR, optical, and fused SAR-optical data, to train and evaluate six ML classifiers. The results demonstrated the effectiveness of moderate-resolution multi-sensor fusion and ML techniques in mapping strategic targets like runways.

Key Takeaways:

  • The study focused on detecting runways and airfield features using open-source Sentinel-1 (SAR) and Sentinel-2 (optical) satellite imagery through three data pathways: PW1 (SAR), PW2 (optical), and PW3 (fused SAR-optical).
  • Six machine learning classifiers were implemented and evaluated, including Random Forest (RF), Extra Trees Classifier (ETC), Light Gradient Boosting Machine (LGBM), Gradient Boosting (GB), Decision Tree (DT), and K-Nearest Neighbors (KNN).
  • The ET classifier in PW3 achieved the highest accuracy (97.98%), followed by LGBM in PW2 (96.24%) and RF in PW1 (73.24%).
  • The research concluded that integrating SAR and optical data enhances classification performance, and RF offers a balance between accuracy and training efficiency.
  • The study highlights the practical implications for defence, disaster response, and remote sensing applications.
  • The research has been peer-reviewed and published in Earth Science Informatics, 2025;18(4).

Statistics:

  • Accuracy of the ET classifier in PW3: 97.98%
  • Accuracy of the LGBM classifier in PW2: 96.24%
  • Accuracy of the RF classifier in PW1: 73.24%
  • Training time of different classifiers:

+ ETC in PW3: 3.05 minutes

+ LGBM in PW2: 2.37 minutes

+ RF in PW1: 1.20 minutes

  • Area Under Curve (AUC) for ETC in PW3: 0.98

Sources:

  • "runway and Airfield Associated Feature Detection From Optical and Sar Data Using Machine Learning Algorithms." Earth Science Informatics, 2025;18(4).
  • "Additional information may be obtained by contacting C. Jeganathan, Birla Institute of Technology Bit, Dept Remote Sensing, Ranchi 835215, Jharkhand, India."