Aggressive behavior recognition and welfare monitoring in yellow-feathered broilers using FCTR and wearable identity tags

Abstract

Aggressive behavior and individual identification in chickens have long attracted widespread attention in animal welfare farming and scientific genetic breeding. Existing methods predominantly rely on manual observation, which is limited by subjectivity and slow response times. The matching of chicken identity with behavior requires considerable human and material resources. To address these challenges, we propose a Fast Chicken aggressive behavior recognition model based on TRansformer (FCTR) model and introduce a wearable identity tag for chickens. FCTR demonstrates robust recognition performance on the free-range Yellow-Feathered Broiler dataset and establishes an identity matching verification method, refining behavioral quantification analysis at the individual level for precise farming. To evaluate this approach, the ChickenFight-2024 dataset was collected and constructed. Multiple experiments confirm that the method can effectively identify both chicken identities and aggressive behaviors using video surveillance images. The proposed model achieved mAP values of 89.81%, 85.76%, 90.14%, 93.19%, and 87.27% for fight, tread, peck, eat, and drink behaviors, respectively, with an mAP of 77.39% for identity information. The identity matching verification method achieved a 94.88% matching rate, highlighting the model’s significant potential for application in commercial farming scenarios and offering new insights and solutions for efficient genetic breeding.