[CVPR 2024 AgVision Workshop] VisTA-SR: Improving the Accuracy and Resolution of Low-Cost Thermal Imaging Cameras for Agriculture
I am pleased to share that our research, “VisTA-SR: Improving the Accuracy and Resolution of Low-Cost Thermal Imaging Cameras for Agriculture”, was selected for an Oral Presentation at the CVPRW 2024 AgVision Workshop (Workshop on Vision4Ag).
This work addresses the critical need for high-resolution thermal imaging in agriculture by leveraging multimodal deep learning to enhance low-cost sensors.
- Title: VisTA-SR: Improving the Accuracy and Resolution of Low-Cost Thermal Imaging Cameras for Agriculture
- Authors: Heesup Yun, Sassoum Lo, Christine H. Diepenbrock, Brian N. Bailey, and J. Mason Earles
- Venue: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024
- Presentation Type: Oral Presentation
Research Summary
Thermal imaging is a vital tool for measuring plant traits such as water stress and stomatal conductance. However, high-resolution thermal cameras are often cost-prohibitive for large-scale agricultural deployment. VisTA-SR is a deep learning-based framework designed to bridge this gap by improving both the temperature accuracy and image quality of consumer-grade thermal sensors.
Key technical contributions include:
- Multimodal Fusion: Leverages high-resolution RGB images as a reference to guide the super-resolution (SR) process of low-resolution thermal data.
- Dual-Stage Network: Features an alignment stage for spatial consistency followed by a sophisticated SR network to restore sharpness and detail.
- Radiometric Calibration: Includes a complete pipeline for temperature accuracy enhancement, validated through field experiments in garbanzo bean crops.
- Improved Phenotyping: Enables precise, organ-level (leaf, stem, fruit) analysis using cameras that cost a fraction of industrial alternatives.
My Contribution:
- Lead author; developed the model architecture, conducted field data collection/validation, and performed the comparative analysis.
Links & Resources
- Paper (CVF Open Access): Full PDF
- Arxiv: 2405.19413
- HTML Version: CVPRW 2024 Paper