Computational models automatically identify sow lameness
Brazilian researchers created a repository of images and videos of sows with different locomotion scores. After, they developed a computer vision model for automatic detection of sow lameness using deep learning. They published their findings in Scientific Reports.
Sow lameness is a common condition affecting mobility and causing pain, discomfort, and poor welfare. Early lameness detection is essential for rapid interventions, and effective treatment, thus improving animal welfare. Therefore, there is an increasing need for automated and non-invasive systems to detect early-stage lameness and improve health and welfare.
Details of the study
The researchers used 500 sows for this trial that lasted for 9 days. They recorded 2D images of sows in locomotion with different lameness scores to create the video image repository. They filed a total of 1.207 videos of lateral and dorsal views of the entire length of each sow. The video processing included each sow entering the corridor and then passing through the corridor. Locomotion experts categorised videos using the Locomotion Score System developed by Zinpro Corporation ranging from no signs of lameness to severe lameness. Then, they created and tested computational models from this annotated repository using a deep learning-based animal pose tracking framework.
Deep learning pose detection model
The researchers created a data repository with lateral and dorsal video images separated by sow locomotion scores with 75% to 100% evaluation confidence. The deep learning pose detection model accurately tracked 6 lateral view and 10 dorsal view skeleton key points with average precisions values of 0.90 and 0.72, respectively. The results of this trial can be used to study the kinematic report of pose detections over time to relate it to the locomotion score assessed by the experts.
The authors concluded that the repository of 2D video images with different locomotion scores contributed to the development of a deep learning pose detection model for early lameness detection in sows.
