Tulu et al. (2025) RGB-to-synthetic-thermal image translation using generative AI to support crop water stress assessment

Identification

• Journal: Computers and Electronics in Agriculture
• Year: 2025
• Date: 2025-12-11
• Authors: Boaz B. Tulu, Fitsum T. Teshome, Yiannis Ampatzidis, Changying Li, Willingthon Pavan, Golmar Golmohammadi, Haimanote K. Bayabil
• DOI: 10.1016/j.compag.2025.111273

Research Groups

• Department of Agricultural and Biological Engineering, Tropical Research and Education Center, IFAS, University of Florida, Homestead, FL, USA
• Central Maryland Research and Education Center, University of Maryland College Park, College Park, MD, USA
• Agricultural and Biological Engineering Department, Southwest Florida Research and Education Center, University of Florida, IFAS, Immokalee, FL, USA
• Department of Agricultural and Biological Engineering, University of Florida, Gainesville, FL, USA
• Department of Soil, Water and Ecosystem Sciences, University of Florida, Gainesville, FL, USA

Short Summary

This study developed and evaluated generative AI models to translate standard RGB images into synthetic thermal images for crop water stress assessment. The Pix2PixGAN model demonstrated high correlation (r > 0.95) with measured thermal data and accurately reflected water stress gradients, offering a cost-effective alternative to specialized thermal sensors for irrigation scheduling.

Objective

• To develop and evaluate generative adversarial network (GAN) models for translating RGB images into synthetic thermal images to support crop water stress assessment and irrigation scheduling.

Study Configuration

• Spatial Scale: 32 experimental plots of sweet corn and green beans. UAV-based orthomosaic RGB and thermal maps were generated.
• Temporal Scale: Three growing seasons, from 2020 to 2023.

Methodology and Data

• Models used: Pix2PixGAN and CycleGAN (both image-to-image translation generative adversarial network deep learning models).
• Data sources:
  • UAV-based RGB and thermal images collected using a Zenmuse XT2 sensor.
  • Data collected from 32 experimental plots of sweet corn and green beans subjected to one full and three deficit irrigation treatments, replicated four times.
  • A total of 3,400 UAV images were collected over three seasons.
  • Image processing performed using Pix4D software to generate orthomosaic RGB and thermal maps, spatially aligned using ground control points (GCPs).
  • Data split: 80% for training and 20% for testing.
  • Image quality evaluation metrics: correlation coefficients (r), mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM).
  • Crop Water Stress Index (CWSI) values computed from both measured and generated thermal imageries.

Main Results

• The Pix2PixGAN model showed a strong correlation (r > 0.95) between generated synthetic thermal canopy temperatures and measured data from a thermal camera.
• Pix2PixGAN outperformed CycleGAN in terms of lower MSE (5.63 vs. 7.09) and higher PSNR (42.98 vs. 40.56).
• CycleGAN had a slightly higher SSIM (0.44) compared to Pix2PixGAN (0.31).
• CWSI values derived from the generated thermal images accurately reflected expected water stress gradients, with the highest CWSI observed in deficit irrigation treatments compared to full irrigation.
• The study demonstrates that RGB-to-synthetic-thermal image translation using GAN models can effectively support crop water stress assessment and irrigation scheduling.

Contributions

• Proposes and validates a novel approach using generative AI (specifically Pix2PixGAN) to synthesize thermal images from standard RGB imagery, addressing the high cost of specialized thermal sensors.
• Provides a practical and cost-effective method for monitoring crop water stress and informing irrigation scheduling using readily available UAV-based RGB data.
• Demonstrates the high accuracy and practical applicability of generated thermal images for calculating the Crop Water Stress Index (CWSI) and reflecting actual water stress conditions across different irrigation treatments.

Funding

• Not specified in the provided text.

Citation

@article{Tulu2025RGBtosyntheticthermal,
  author = {Tulu, Boaz B. and Teshome, Fitsum T. and Ampatzidis, Yiannis and Li, Changying and Pavan, Willingthon and Golmohammadi, Golmar and Bayabil, Haimanote K.},
  title = {RGB-to-synthetic-thermal image translation using generative AI to support crop water stress assessment},
  journal = {Computers and Electronics in Agriculture},
  year = {2025},
  doi = {10.1016/j.compag.2025.111273},
  url = {https://doi.org/10.1016/j.compag.2025.111273}
}