🐄 PRISM-CAFO

Prior-conditioned Remote-sensing Infrastructure Segmentation & Mapping for CAFOs

Scalable • Explainable • Infrastructure-aware AI for Livestock Facility Mapping

---

🌎 Motivation

Concentrated Animal Feeding Operations (CAFOs) significantly impact environmental sustainability, water quality, and public health—yet large-scale monitoring of livestock infrastructure remains limited.

PRISM-CAFO introduces an infrastructure-first, explainable remote-sensing pipeline that detects, segments, and reasons over facility infrastructure (barns, lagoons, feedlots) to identify CAFOs at scale.

---

💡 Core Idea

Instead of treating CAFO identification as pure image classification:

Detect infrastructure → segment objects → extract descriptors → classify → explain

This yields improved interpretability and robustness across regions.

---

🧠 Method Overview

Pipeline

1. Infrastructure Detection (YOLO): detect candidate barns/lagoons/feedlots
2. Segmentation (SAM): refine to pixel-accurate masks
3. Descriptor Extraction: counts, areas, shapes, and spatial layout features
4. Multimodal Prediction: fuse visual + structured descriptors
5. Explainability: attribution highlights decisive infrastructure cues

---

🖼 Visual Results

Infrastructure Detection

Segmentation (SAM)

Outputs: Explainable Mapping

---

📂 Repository Structure

PRISM-CAFO/
├── datas/                      # dataset + metadata
├── model/                      # trained weights
├── main.py                     # main pipeline
├── train_yolo.py               # infrastructure detector training
├── train_multiclass_v2.py      # CAFO classifier training
├── runPredict.py               # inference
├── run_sam_on_yolo.ipynb       # segmentation notebook
├── filtering_components.ipynb
└── docs/                       # README visuals

(Structure verified from the GitHub repository file tree.)

---

⚙️ Installation

Create environment

conda create -n prism_cafo python=3.10 -y
conda activate prism_cafo

Install dependencies

pip install torch torchvision
pip install ultralytics
pip install segment-anything
pip install opencv-python matplotlib rasterio tqdm

If you run on HPC: install a CUDA-matching PyTorch build and ensure your CUDA runtime is visible.

---

🚀 Training

Train infrastructure detection model

python train_yolo.py

Train CAFO classifier

python train_multiclass_v2.py

---

🔍 Inference

python runPredict.py

SAM segmentation notebook:

run_sam_on_yolo.ipynb

---

📚 Paper

PRISM-CAFO: Prior-conditioned Remote-sensing Infrastructure Segmentation and Mapping for CAFOs (WACV 2026)
arXiv: https://arxiv.org/abs/2601.11451

---

👥 Authors

Oishee Bintey Hoque • Nibir Chandra Mandal • Kyle Luong • Mandy Wilson • Samarth Swarup • Madhav Marathe • Abhijin Adiga

---

⭐ Citation

@inproceedings{prismcafo2026,
  title={PRISM-CAFO: Prior-conditioned Remote-sensing Infrastructure Segmentation and Mapping for CAFOs},
  author={Hoque, Oishee and Mandal, Nibir and others},
  booktitle={WACV},
  year={2026}
}

---

🔮 Roadmap

• Release pretrained models
• Release benchmark data splits
• Interactive mapping dashboard
• Temporal change detection

---

🙏 Acknowledgements

Developed at the Biocomplexity Institute, University of Virginia.