LAUSR

Abstract Broiler chickens are traditionally weighed by steelyard or platform scale, which is time-consuming and labor-intensive. Broiler chickens usually exhibit stress-related behavior during weighing. The 3D camera-based weighing system for broiler chickens can only weigh the broiler chicken in the monitoring area. Usually, it makes poor weight prediction due to poor segmentation especially when the broiler chicken is flapping its wings. To solve these issues, we developed one simple and low-cost weighing system with high stability and accuracy. A validity value extraction method from dynamic weighing was proposed. Then, an improved amplitude-limiting filtering algorithm and a BP neural networks model were developed to avoid accidental interference. The BP neural networks model used daily weight gain, day-age, average velocity, and the weight data after filtering algorithm as the input layer. The weighing system was tested in a commercial Beijing Fatty Chickens house with Beijing Fatty Chickens. We tested thirteen groups of Beijing Fatty Chickens of different weights, from 500 g to 1800 g in intervals of 100 g, using the three different methods: no filtering algorithm or BP neural networks, only the improved amplitude-limiting filtering algorithm and a hybrid of the improved amplitude-limiting filtering algorithm and BP neural networks. The results showed that the hybrid algorithm had a better performance in minimizing the error, lowering from the original 6% down to 3%. The accurate weight data was transmitted to the remote service platform for further decision-making, such as activity analysis, feeding management, and health alerts.