2 / 2025-03-11 01:01:21

Developing Deep Learning Models for Automated Mating-Related Behavior Detection in Broiler Breeders in Lab Setting

Fertility,Broiler breeder,Mating behavior,Mounting

3. 智慧牧场与动物福利 > （4）智慧牧场应用的案例实践

Mating behavior in broiler breeders plays a crucial role in ensuring egg fertility. It begins with the male mounting the female, followed by tail movements from both to facilitate cloacal contact. These behaviors occur sporadically and rapidly which makes manual observation challenging. This study aimed to develop a vision-based deep learning (DL) model to automatically detect the mounting behavior in broiler breeders and determine its frequency, duration, interval, as well as tail movement incidences. The experiment was conducted with four pens each containing 10 hens (Cobb 500F) and 1 rooster (Cobb MX). Room environment was maintained within the Cobb recommendation. Side and top view cameras were installed at each pen recording throughout the 15-hour daily light period. A DL model, YOLOv8, was trained to detect mounting behavior with a dataset consisting of 1,620 mounting and 800 non-mounting instances. Major welfare indicators, including footpad dermatitis (FPD), gait score (GS) and weight were manually assessed weekly. The YOLOv8 model achieved 91.0% and 90.0% training accuracies in detecting mounting and non-mounting behaviors by roosters, respectively. The roosters exhibited an average mounting frequency of 13.6 ± 2.7 times per day, with 5.3±2.4 seconds mean duration per mounting, and 63.2±81.9 minutes inter-mounting interval. On average, 80.6% of the mounting attempts were carried out with observable rooster tail movements. The peak mounting activity periods were found to be during the 06:00–09:00 and 16:00–20:00 hours. The age (p<0.0001), rooster FPD (p=0.0230) and hen GS (p<0.0001) were strongly affecting the mounting frequencies.