VentVision: A multimodal vision-based system for automated vent-based chick sexing
Marta Veganzones Rodriguez, Thinh Phan, Arthur Francisco de Araújo Fernandes, Vivian P. Breen, Jesus Arango, Michael T. Kidd, Ngan Le · Smart Agricultural Technology · 2026
Chick sexing has been a crucial practice for many years due to the need for early gender identification in poultry production. Traditional methods, such as vent sexing, are highly effective but labor-intensive, requiring expert knowledge and years of training, making it a challenging and time-consuming process. Moreover, skilled professionals are increasingly scarce, making the process difficult to scale. To address this challenge, we proposed VentVision , an automated vision-based classification system that leverages computer vision and deep learning techniques to improve the efficiency and scalability of chick gender classification. Our system follows a four-step pipeline: multi-modality video capture, temporal segment selection, vent region detection, and gender classification, automating vent-based sexing. We evaluated our approach using RGB imaging, infrared (IR) imaging, and a multimodal RGB-IR feature-level fusion framework based on a dual-stream architecture. The multimodal model achieved an accuracy of 98.09%, outperforming single-modality approaches. These results demonstrate that VentVision achieves expert-level chick sexing performance while eliminating the need for years of specialized training, with manual handling limited to vent exposure, thereby simplifying the sexing workflow.