RemoteSense: Burn-scar-based crop residue burning detection and alerting with a fine-tuned vision language model on Sentinel-2 imagery

Author

Sahil Patil, Amol Powar, Sarvesh Parab, Prof. Revti Jadhav


Abstract

 Crop residue burning in northern India causes recurring air-quality crises, yet active-fire sensors (MODIS, VIIRS) miss most events: their fixed overpass times are increasingly avoided by farmers, and a CEEW study found VIIRS detected only 7 of 169 burnt fields (4.1%) in Sangrur over two days. Because burn scars persist long after flames die, we propose detecting scar evidence rather than the fire itself. We present RemoteSense, an end-to-end platform that (i) fine-tunes a 450M-parameter vision language model (LFM2.5-VL-450M) with LoRA to read paired RGB and SWIR Sentinel-2 composites and return a structured JSON burn assessment; (ii) corroborates detections with live NASA FIRMS hotspots; (iii) dispatches severity-routed, geo-tagged alerts via SMS, email, web push, and webhooks; and (iv) exports the model to quantized ONNX for browser and edge inference. On 178 clear-sky, district-disjoint test tiles, fine-tuning raises mean five-field accuracy from 21.9% to 74.8% and severity accuracy from 11.8% to 53.9%; binary burn detection reaches 73.6% versus a 65.2% majority-class baseline. These results are preliminary: labels are semi-automatically generated, the test set is single-season, and no comparison with spectral-index or CNN baselines is yet available.


Keywords

vision language model, crop residue burning, Sentinel-2, burn scar detection, LoRA fine-tuning, edge AI, alert system



Full Text:

Download Paper PDF


References


  • Anand, A., Imasu, R., Dhaka, S. K., & Patra, P. K. (2025). Domain adaptation and fine-tuning of a deep learning segmentation model of small agricultural burn area detection using high-resolution Sentinel-2 observations: A case study of Punjab, India. Remote Sensing, 17(6), 974.
  • Ayanzadeh, A., Dixit, P., Kamal, S., & Halem, M. (2026). WildfireVLM: AI-powered analysis for early wildfire detection and risk assessment using satellite imagery (arXiv:2602.13305). arXiv.
  • (2025). How can satellite-borne remote sensing capture crop burning incidents? Council on Energy, Environment and Water, New Delhi, India. https://www.ceew.in/publications/how-can-satellite-borne-remote-sensing-capture-crop-burning-and-farm-fires
  • Deshpande, M. V., Pillai, D., & Jain, M. (2022). Detecting and quantifying residue burning in smallholder systems: An integrated approach using Sentinel-2 data. International Journal of Applied Earth Observation and Geoinformation, 108, 102761.
  • Dettmers, T., Pagnoni, A., Holtzman, A., & Zettlemoyer, L. (2023). QLoRA: Efficient finetuning of quantized LLMs. In Advances in Neural Information Processing Systems (NeurIPS).
  • DPhi Tech. (n.d.). SimSat: Satellite imagery API. https://dphi.tech
  • Drusch, M., et al. (2012). Sentinel-2: ESA’s optical high-resolution mission for GMES operational services. Remote Sensing of Environment, 120, 25–36.
  • European Space Agency. (n.d.-a). Φ-sat programme: Φsat-2. ESA Φ-lab. https://philab.esa.int/flagship-programmes/phi-sats-programme/
  • European Space Agency. (n.d.-b). Copernicus Data Space Ecosystem. https://dataspace.copernicus.eu
  • Filipponi, F. (2018). BAIS2: Burned area index for Sentinel-2. Proceedings, 2(7), 364.
  • Hu, E. J., et al. (2022). LoRA: Low-rank adaptation of large language models. In Proceedings of the International Conference on Learning Representations (ICLR).
  • Key, C. H., & Benson, N. C. (2006). Landscape assessment: Ground measure of severity, the composite burn index. In FIREMON: Fire Effects Monitoring and Inventory System(USDA Forest Service General Technical Report RMRS-GTR-164-CD).
  • Knopp, L., Wieland, M., Rättich, M., & Martinis, S. (2020). A deep learning approach for burned area segmentation with Sentinel-2 data. Remote Sensing, 12(15), 2422.
  • Kuckreja, K., Danish, M. S., Naseer, M., Das, A., Khan, S., & Khan, F. S. (2024). GeoChat: Grounded large vision-language model for remote sensing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 27831–27840).
  • Li, X., Wen, C., Hu, Y., Yuan, Z., & Zhu, X. X. (2024). Vision-language models in remote sensing: Current progress and future trends. IEEE Geoscience and Remote Sensing Magazine, 12(2), 32–66.
  • Liquid AI. (2026). LFM2.5-VL-450M [Model card]. https://huggingface.co/LiquidAI/LFM2.5-VL-450M
  • Miller, J. D., & Thode, A. E. (2007). Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sensing of Environment, 109(1), 66–80.
  • (n.d.). Fire Information for Resource Management System (FIRMS). https://firms.modaps.eosdis.nasa.gov
  • Ramo, R., et al. (2021). African burned area and fire carbon emissions are strongly impacted by small fires undetected by coarse resolution satellite data. Proceedings of the National Academy of Sciences, 118(9), e2011160118.
  • (2026, September 28). India’s TakeMe2Space to launch orbital computing satellite on SpaceX rocket.
  • Schroeder, W., Oliva, P., Giglio, L., & Csiszar, I. A. (2014). The new VIIRS 375 m active fire detection data product: Algorithm description and initial assessment. Remote Sensing of Environment, 143, 85–96.
  • Seidel, L., et al. (2025). Advancing early wildfire detection: Integration of vision language models with unmanned aerial vehicle remote sensing for enhanced situational awareness. Drones, 9(5).
  • Soni, S., et al. (2025). EarthDial: Turning multi-sensory Earth observations to interactive dialogues. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(pp. 14303–14313).
  • (2025). Are farm fires going uncounted? ISRO study suggests farmers have figured out a way to evade satellites. https://theprint.in/environment/are-farm-fires-going-uncounted-isro-study-suggests-farmers-have-figured-out-a-way-to-evade-satellites/2797583/
  • The Tribune. (2025, December). Satellites miss majority of stubble fires, Delhi air pollution underestimated: Report [On the iFOREST Stubble Burning Status Report 2025]. https://www.tribuneindia.com/news/delhi/satellites-miss-majority-of-stubble-fires-delhi-air-pollution-underestimated-report
  • Trigg, S., & Flasse, S. (2001). An evaluation of different bi-spectral spaces for discriminating burned shrub-savannah. International Journal of Remote Sensing, 22(13), 2641–2647.
  • Walker, K., et al. (2022). Detecting crop burning in India using satellite data (arXiv:2209.10148). arXiv.
  • Wang, F., et al. (2025). GeoLLaVA-8K: Scaling remote-sensing multimodal large language models to 8K resolution. In Advances in Neural Information Processing Systems (NeurIPS)(Vol. 38).

Share your valuable work from Social Media Buttons