- Challenges
Embedded sensors and computer vision, enhanced by artificial intelligence (AI), enable automated tracking of animal locations and behaviours. However, ensuring the robustness of AI algorithms in tracking animal access to outdoor environments and generating animal welfare Key Performance Indicators (KPIs) requires rigorous testing in diverse environmental conditions before widespread adoption of these solutions.
- Solution
Case study C addresses these challenges with multi-farm datasets utilizing two distinct technologies: computer vision for pigs and GPS sensors for dairy cows, specifically focusing on their access to grassland. i) Utilizing animal-level data from 20 dairy farms with varying characteristics (e.g., breed, housing, environment) from another project, a novel tool called the Smart Labeling Loop will be developed as a web application. This tool facilitates the manual labelling of diverse behaviour data to train neural networks, ensuring robust algorithm development. ii) Video data from 13 pig farms, linked to a breeding program, will be utilized to evaluate the performance of AI models against a reference model.
- Outcomes
Digi4Live aims to demonstrate the performance of different technologies and algorithms across diverse conditions, and to identify their limitations. Additionally, it seeks to develop prediction models for outdoor access time, which will be compared against observational and RFID data. Digi4Live will develop general guidelines for training robust AI models based on Internet of Things (IoT) sensors and computer vision.
- Replicability
The method employed by CS-C is designed replicable to various applications, providing valuable insights on the reliability of AI in different settings.