Work

A computational approach for automated assessment of poultry barns with field validation

Year: 2025

Type: article

Abstract: Many poultry farms in the US are smaller in size than the threshold set by Concentrated Animal Feeding Operations (CAFOs), exempting them from having operating permits and state registration. This lack of oversight makes it challenging for producers, state agencies, and environmental groups to identify the locations and capacities of these facilities. The aim of our study was to develop a deep learning-based automated method to detect poultry barns within the five-county region of southwest Missouri and estimate their capacity. The study found that only 187 poultry operations were reported in CAFOs dataset, but 491 additional poultry farm structures were identified in the study area. Deep learning models were employed on aerial imagery to automatically detect poultry barns and estimate their area. A semi-automated approach was used to estimate the barn’s capacity at the county level. The mean accuracy of the selected model reached 91.96% in detecting number of barns and 89.27% in estimating areas. The model detected 1,158 unregistered barns in the five-county region, with an estimated capacity of 33.26 million broiler-equivalent birds. The study highlighted the need to identify the unregistered poultry operations due to biosecurity, manure management, and environmental concerns. A web-based tool was developed to allow users to navigate through the results of the tools within the five-county study area for demonstration and visualization purposes. This computational tool can facilitate the accurate identification of poultry farms, providing insights into their characteristics, spatial distribution, and production capacities, thereby supporting more informed decision-making and improved farm management practices.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2025.101390