PRISM-CAFO: Prior-conditioned Remote-sensing Infrastructure Segmentation and Mapping for CAFOs

TL;DR

PRISM-CAFO tackles the challenge of scalable, interpretable CAFO mapping from aerial imagery by coupling an infrastructure-focused detection stage with SAM2-based refinement, geometric filtering, and a priors-informed, attention-based classifier. The method jointly leverages explicit infrastructure descriptors and domain statistics to produce CAFO type predictions with mask-level attributions, enabling interpretable linking of decisions to barns, lagoons, and other components. Across a large, multi-state NAIP-derived dataset, PRISM-CAFO achieves state-of-the-art performance, with notable gains under distribution shifts and strong ablation results demonstrating the value of each component. The approach offers a general, scalable template for infrastructure-conditioned segmentation and classification in remotely sensed imagery, with potential applications in environmental monitoring, epidemiology, and policy analysis.

Abstract

Large-scale livestock operations pose significant risks to human health and the environment, while also being vulnerable to threats such as infectious diseases and extreme weather events. As the number of such operations continues to grow, accurate and scalable mapping has become increasingly important. In this work, we present an infrastructure-first, explainable pipeline for identifying and characterizing Concentrated Animal Feeding Operations (CAFOs) from aerial and satellite imagery. Our method (1) detects candidate infrastructure (e.g., barns, feedlots, manure lagoons, silos) with a domain-tuned YOLOv8 detector, then derives SAM2 masks from these boxes and filters component-specific criteria, (2) extracts structured descriptors (e.g., counts, areas, orientations, and spatial relations) and fuses them with deep visual features using a lightweight spatial cross-attention classifier, and (3) outputs both CAFO type predictions and mask-level attributions that link decisions to visible infrastructure. Through comprehensive evaluation, we show that our approach achieves state-of-the-art performance, with Swin-B+PRISM-CAFO surpassing the best performing baseline by up to 15\%. Beyond strong predictive performance across diverse U.S. regions, we run systematic gradient--activation analyses that quantify the impact of domain priors and show ho

Figures (5)

• Figure 1: PRISM-CAFO's candidate-infrastructure-focused approach not only leads to better identification of CAFOs, but also captures key infrastructure in the process with higher confidence scores (based on GradCAM selvaraju2017grad).
• Figure 2: Qualitative comparison of existing object identification approaches with our domain-specific module.
• Figure 3: The schematic of the PRISM-CAFO framework. Refined YOLO and SAM2 masks form synthetic priors that modulate visual embeddings via masked-guided spatial attention, producing prior-aware features for accurate CAFO classification.
• Figure 4: Feature importance scores using gradient-activation analysis ($|\nabla \cdot \text{act}|$) for livestock classification. Barn area dominates (0.74), followed by domain priors (0.15--0.47) and barn-pond proximity (0.27). Blue: extracted features, orange: prior knowledge, green: proximity measures.
• Figure 5: Probability-drop ($\Delta_k$) analysis of infrastructure masks. Barns dominate across all classes (0.45 swine, 0.33 beef, and 0.25 dairy and poultry), with manure ponds adding signals for swine and beef. Other components remain comparatively small.