← Back to SinkAlert Dashboard

📚 Academic Literature Review

35+ papers reviewed across 6 research domains. SinkAlert builds on proven science — not reinvention.

🔍 Key Finding: No existing system combines all three data modalities (Satellite InSAR + Dashcam CV + Environmental ML) for road collapse prediction. SinkAlert's three-source fusion is genuinely novel.
🛰️ InSAR + ML (13 papers) 📡 SBAS-InSAR / PSI (8 papers) 🚗 YOLO Road Damage (12 papers) 🧠 XGBoost Geohazard (11 papers) 🇹🇭 Thailand Research (4 papers) 💻 GitHub Repos (15+ repos) 📦 Key Datasets 📖 Citation Strategy

🛰️ 1. InSAR + ML for Infrastructure Monitoring

Foundational Papers

#PaperYearKey Contribution
1.1 Sinkhole Scanner: A New Method to Detect Sinkhole-Related Spatio-Temporal Patterns in InSAR Deformation Time Series
DOI: 10.3390/rs13152906		 2021 First method combining InSAR time series with ML for automated sinkhole detection. Closest prior art to SinkAlert.
1.2 InSAR Time Series Analysis for Sinkhole Detection using Deep Learning (PhD Thesis)
DOI: 10.3990/1.9789036557283		 2023 CNN-LSTM + UNet for InSAR sinkhole detection. Open-source code available.
1.9 Urban Ground Subsidence Monitoring and Prediction Using Time-Series InSAR and ML (Tianjin case study)
10.21203/rs.3.rs-4370214/v1		 2024 InSAR + XGBoost, RF, LSTM for urban subsidence. Highly relevant architecture.
1.10 Optimization of Land Subsidence Prediction Features Based on ML and SHAP with Sentinel-1 InSAR
10.21203/rs.3.rs-3880879/v1		 2024 Directly relevant: Feature optimization with SHAP + InSAR + ML for subsidence prediction.
1.12 InSARTrac: Combining Computer Vision and Terrestrial InSAR for 3D Displacement Monitoring
10.3390/rs15082031		 2023 Pioneering CV + InSAR fusion. Proves combining both modalities is feasible.

📡 2. SBAS-InSAR / PSI Techniques

Key Software for SinkAlert: Use LiCSBAS (⭐279, Python) for automated SBAS-InSAR processing of Sentinel-1 data. Open-source, well-documented, designed for COMET LiCSAR products covering Thailand.
SoftwareStarsDescription
MintPy⭐798Miami InSAR time-series. Production-grade.
PyGMTSAR⭐583Powerful Python InSAR. SBAS + PSI.
LiCSBAS ✅⭐279SBAS using COMET LiCSAR Sentinel-1 products. Recommended for SinkAlert.
InSARHub⭐160End-to-end InSAR with GUI.

🚗 3. YOLO-Based Road Damage Detection

#PaperYearKey Contribution
3.1 RDD2022: A Multi-National Image Dataset for Automatic Road Damage Detection
10.1002/gdj3.260 / arXiv:2209.08538		 2024 THE benchmark dataset. 47,420 images from 6 countries. Thailand-relevant damage types.
3.2 Optimizing YOLO Architectures for Road Damage Detection: YOLOv7 to YOLOv10
arXiv:2410.08409		 2024 YOLOv8 best balance of speed/accuracy on RDD dataset.
3.4 Intelligent Road Crack Detection Based on Improved YOLOv8
arXiv:2504.13208		 2025 YOLOv8 + attention mechanisms. SOTA results.

Implementable GitHub Repos

RepoStarsDescription
FarzadNekouee/YOLOv8_Pothole_Segmentation_Road_Damage_Assessment⭐55YOLOv8-seg for pothole detection + severity. Working code.
KaikePing/RoadDamageYOLO⭐10YOLOv5/v8/v11 comparison on RDD2022.
bharath-alavala123/Automated-Pavement-Distress-Detection-using-YOLOv8⭐4YOLOv8-medium for pavement distress.

🧠 4. XGBoost for Geohazard Prediction

Recommendation: XGBoost is the most proven algorithm for geohazard prediction in the literature. Use SHAP for model interpretability — crucial for judge credibility.
#PaperKey Insight
4.6 Slope Unit-Based Geohazard Susceptibility: SHAP + InSAR Deformation
10.1016/j.asr.2025.03.034		 Most relevant: Combines InSAR + XGBoost/LightGBM + SHAP for geohazard assessment.
4.10 Urban Ground Subsidence: Time-Series InSAR and ML (XGBoost, RF, LSTM)
10.21203/rs.3.rs-4370214/v1		 XGBoost excelled for feature-rich InSAR data.
4.11 SHAP-Based Feature Optimization for Subsidence with Sentinel-1 InSAR
10.21203/rs.3.rs-3880879/v1		 Methodology directly applicable to SinkAlert.

🇹🇭 5. Thailand & Bangkok-Specific Research

#PaperYearKey Finding
5.1 InSAR Time-Series Analysis of Land Subsidence in Bangkok
10.1080/01431161.2012.756596		 2013 Foundational paper. First comprehensive Bangkok InSAR analysis. Subsidence up to 30 mm/yr.
5.2 Land Subsidence in Bangkok: Causes and Long-Term Trends Using InSAR and ML
10.2139/ssrn.4760676		 2024 Directly relevant. InSAR + ML for Bangkok subsidence trends.
5.3 Evolution Pattern of Land Subsidence in Bangkapi, Bangkok
10.59796/jcst.v14n3.2024.49		 2024 Recent InSAR analysis of specific Bangkok district.
Gap Confirmed: Bangkok has severe ongoing subsidence (10–30 mm/yr). No operational road collapse prediction system exists in Thailand or ASEAN. SinkAlert fills a clear gap.

📦 6. Key Datasets

DatasetSizeUse in SinkAlert
RDD202247,420 imagesTrain YOLO for dashcam CV
Sentinel-1 (ESA)Global, freeInSAR deformation (6-12 day revisit)
COMET LiCSARGlobalPre-processed interferograms via LiCSBAS
CHIRPS Rainfall1981-presentEnvironmental feature for XGBoost
SMAP Soil MoistureGlobal, 9kmSoil moisture feature
OpenStreetMapGlobalRoad network segmentation

📖 7. Citation Strategy for Judges

Proven Techniques We're Building On

  1. SBAS-InSAR → Morishita et al. (2020) — LiCSBAS
  2. YOLOv8 road damage → Arya et al. (2024) — RDD2022 + Zuo et al. (2025)
  3. XGBoost geohazard → Badola et al. (2023) + Wang et al. (2025) — InSAR+ML+SHAP
  4. Sinkhole detection → Kulshrestha et al. (2021) — Sinkhole Scanner
  5. Bangkok context → Aobpaet et al. (2013) + Ahmed et al. (2024)
  6. CV+InSAR fusion → Zambanini et al. (2023) — InSARTrac

Novel Contributions (What SinkAlert Adds)

  1. ✅ First system combining all 3 data sources (InSAR + CV + environmental) for road collapse
  2. ✅ Operational 100m-segment risk index — actionable for road authorities
  3. ✅ ASEAN/Thailand-specific deployment — filling a real gap
  4. ✅ SHAP-based explainability — transparent risk scoring
  5. ✅ Hybrid update frequency — real-time dashcam + periodic satellite