Weight prediction method for individual live chickens based on single-view point cloud information

The research of poultry phenotypic measurement is an important area of Precision Livestock Farming (PLF). With continuously increasing standards for farm animal production, accurately obtaining various phenotypic parameters of poultry is particularly important. Body weight is today the most crucial phenotypic parameter for poultry production. It serves as a primary indicator for monitoring growth and selecting breeding stock. Traditionally, weighing poultry has been done using platform scales. However, the accuracy of these scales is affected by the chickens’ movements. To address this challenge, this paper proposes a method for predicting live chicken weight based on single-view point cloud data derived from a depth camera. This camera captures a top-view point cloud of the chickens’ back, and with an improved deep learning model based on PointNet++ for weight prediction using this point cloud information as a single input feature. On the test dataset, a mean absolute error of 95 g (6.66 % relative error) was found in the weight prediction, with the Pearson correlation coefficient with the observed weights being 0.8817. These results indicate that the proposed method performs well in predicting live chicken weight. Besides, the cross-validation experiment showed that the results of weight prediction, using single breed data set, were similar to those of a mixed data set (MAE 92 g, MAPE 6.81 % and Pearson correlation coefficient 0.7997 for Huainan partridge chicken; MAE 95 g, MAPE 7.35 % and Pearson correlation coefficient 0.8265 for Huxu chicken). This study demonstrates that the weight prediction model effectively captures feature differences among different breeds, yielding accurate results. As the first to combine point cloud data with deep learning for live chicken weight prediction, it conducts comparative analyses of various models and discusses limiting factors and improvement strategies. Additionally, the single-viewpoint data acquisition method offers valuable insights for practical applications in commercial poultry farms.