Automated on-site broiler live weight estimation through YOLO-based segmentation

Abstract

Broiler weighing is essential in poultry production for growth monitoring, feed management, health detection, and meeting market requirements. Traditional weighing methods, which use electronic platform weighers, can stress broilers and may not capture accurate weight data, particularly heavy broilers. To overcome these limitations, this study proposes a camera-based weighing approach that relies on morphological changes in different growth stages of broilers rather than body dimensions. The study utilizes YOLO version 8, a deep learning-based network segmentation technique, for precise broiler segmentation, significantly improving weight accuracy in complex environments. The YOLOv8 architecture builds a model that demonstrates improved and trustworthy results in broiler weight prediction, achieving a mean average precision across a range of intersection over union thresholds from 50 % to 95 % of 0.829. By accurately estimating broiler weights based on their morphological features, the developed trained YOLOv8 model eliminates the need for measuring their dimensions or sizes, making the process efficient and convenient.