Advancing Poultry Farming Efficiency Through YOLOv5 and Image Regression-Based Broiler Weight Estimation

The precise calculation of broiler chicken weight is crucial for the efficient management of the farm and optimization of production. Conventional weighing methods are labor-intensive and can induce stress in the poultry, highlighting the need for non-invasive and precise measurement techniques. The system developed in this paper is a method designed for Cobb500 broilers, whereby an innovative image analysis approach yields weight estimates. The system has used two models: YOLOv5 and image regression. The YOLOv5 model used in this study was trained with 3920 annotated images and 490 test sets, while the image regression model had 2240 labeled data with a 280 test set. Evaluation of the models was done on a separate farm to ensure unbiased results, which yielded a score of 0.00053kg MSE, 0.02303kg RMSE, and an R2 of 0.73 on single broiler detections, and 0.00026kg MSE, 0.01625kg RMSE, and 0.77 R2 on multiple broiler detections. These results outperformed previous similar studies in terms of MSE and RMSE, demonstrating the system's enhanced accuracy and reliability in broiler weight estimation.