Livestock Images and Behavior

The development of AI tools in agriculture requires the creation of specific databases to train algorithms deployed in the field. Particularly, image processing requires annotated image datasets where all relevant information is manually identified. These data are used to train algorithms for the automatic extraction of this information from images acquired in similar contexts. Despite the multiple possibilities offered by image analysis methods, their adoption in livestock farming is slow due to the lack of quality databases. More precisely, database characterized by a wide diversity of environments and conditions, along with a large number of precisely annotated images. The objective of the LIB project is to address this gap by establishing such a database, leveraging the diversity of experimental farms at INRAE, and hopefully over the world, to support future developments in AI-based livestock farming management.

The database will primarily consist of images of livestock animals raised in group. Each image will be manually annotated to indicate the position of the animals body and head in the image, specify the species, and relevant behavioral traits. This database could then be used to train a neural network for the estimation of animal locations on images, such as Yolo, as well as classification CNNs to estimate animal behavior. These neural networks will be available at the end of the project for use in various applications related to image analysis in livestock farming.

The database, along with the developed neural networks, will enable the creation of tools aimed at estimating various behavioral variables of interest, such as activity level, postural time budgets, the use of areas of interest, or interactions between conspecifics. These variables will then be used to assess and characterize the health and well-being of animals over time. In particular, these neural networks can be directly integrated into video surveillance systems to facilitate real-time animal management or to measure relevant behavioral traits in the context of genetic selection or well-being monitoring, for examples.

Finally, it is an excellent opportunity to demonstrate that the livestock community in this field is capable of coming together to overcome a common obstacle !

For more information, please contact us !