Live weight prediction from top view images of Akkaraman male lambs using digital image processing and machine learning methods

The rapid and accurate determination of live weight in sheep farming holds great importance for effective flock management. Traditional live weight measurement methods are time-consuming and pose various challenges for both animals and workers. This study aimed to estimate the live weight of sheep using digital image processing and machine learning methods based on dorsal images. The research utilized dorsal images and corresponding live weight data of 38 Akkaraman male lambs, raised at the Livestock Application and Research Center of Van Yüzüncü Yıl University, collected at three different ages (3, 4, and 5 months). High-resolution images obtained from video recordings were processed to extract morphological features, forming the dataset. Random Forest (RF), Support Vector Machines (SVM), Multivariate Adaptive Regression Splines (MARS), and Extreme Gradient Boosting (XGBoost) algorithms were modeled and their performances were compared. The prediction performances of the models were evaluated on test data using statistical metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²). The MARS model exhibited the highest prediction performance (R²=0.94, MAE=1.54 kg, RMSE=2.14 kg), followed by RF (R²=0.92, MAE=1.88 kg, RMSE=2.60 kg) and XGBoost (R²=0.91, MAE=1.96 kg, RMSE=2.83 kg). The findings demonstrate that image processing and machine learning techniques can serve as reliable and practical alternatives for live weight estimation in sheep farming. This study represents a significant step toward applying technological advancements in livestock farming and achieving digital transformation in flock management.