Title: CAFOSat: A High-Resolution, Infrastructure-Aware Dataset for Mapping Concentrated Animal Feeding Operations

Abstract:

Concentrated Animal Feeding Operations (CAFOs) are pivotal to agricultural output, yet they simultaneously raise significant issues regarding environmental protection, public health, and disease monitoring. Accurately mapping these facilities from remote sensing data is difficult, largely due to inconsistent annotations, fragmented inventories, noisy location data, and the complex, varied layouts of CAFO infrastructure. To address these hurdles, we present CAFOSat, a comprehensive, strongly annotated dataset designed for CAFO detection across the United States, with a specific focus on infrastructure awareness.

CAFOSat merges high-resolution imagery from the National Agriculture Imagery Program (NAIP) with multi-state CAFO inventory records. We enhance the precision of weak geolocation records by employing a human-in-the-loop workflow that integrates AI-assisted annotation, GradCAM-based localization techniques, and geometric clustering. To ensure high data quality, we utilize spatial exclusion constraints and land-cover-guided sampling to generate challenging negative samples. Furthermore, we provide detailed, infrastructure-level labels—such as manure ponds, barns, and grazing areas—which are verified manually.

The final dataset comprises over 45,000 image patches covering four primary CAFO categories across 20 states. Our benchmarks, which evaluate a wide range of models including convolutional networks, transformers, and vision-language architectures, highlight the effectiveness of curated negative samples and refined annotations in boosting CAFO classification and generalization capabilities. Additionally, we propose a synthetic augmentation strategy that creates infrastructure-aware variations to enhance training diversity and model robustness against distribution shifts. CAFOSat establishes a large-scale benchmark to advance the field of agricultural monitoring, facilitating more accurate CAFO mapping from high-resolution remote sensing imagery.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC