In traditional poultry farms, operators are required to manually inspect chicken houses to investigate whether the chickens exhibit abnormal behaviors. Although this method seems simple, it is labor-intensive, time-consuming, inefficient, and subjective. Furthermore, the poultry industry is also facing major problems of labor shortage and aging.
In order to solve the aforementioned problems, this study proposes the use of deep learning to objectively and accurately monitor the chickens in the chicken houses in real time, and analyze and report behavioral and environmental data, while using the least amount of labor and time. Our proposed solution is expected to not only solve the major challenges faced by the poultry industry today, but also promote a smarter breeding environment with more accurate, efficient and effective use of resources.
In this study, we built an overhead image capture system consisting of a camera array, a cloud server, a deep learning model, and an interactive webpage. The camera array is installed on the roof of the chicken house, and the collected images are sent to the cloud server. The images are subsequently passed to an object detection model, which provides information about the positions of the chickens in the chicken house, followed by a trajectory tracking algorithm to analyze the activity of each individual chicken, as well as the overall distribution of the chickens. Finally, the results are displayed graphically on the interactive webpage, allowing the user to monitor the condition of the chickens in real time.
At present, the proposed deep learning model is able to quickly and accurately mark the positions of chickens in an image, which is conducive to the subsequent analysis of chicken activity and distribution.
Comments