Detecting estrus in sows using a robotic imaging system and neural networks
Loading...
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
The global demand for pork products is expected to grow further with the growth of the population. Traditional swine farming presents disadvantages such as low breeding efficiency and high labor cost. Estrus detection is crucial to reproductive management of sows. Due to the longevity of the sperm and eggs, achieving timely and accurate estrus detection can lead to higher conception rates, larger litter size, and lower non-productive days. However, for decades, estrus detection still relies on manual examination of the sows, and the effectiveness of the current approach has plenty of room for improvement. As the swine farming operation transitions into large-scale, a more effective solution for estrus detection is needed. The goal of this research was to develop a robotic imaging system leveraging advanced technologies in remote sensing and artificial intelligence for evaluating key traits that can be used to accurately detect the onset of estrus in sows. To achieve this goal, this research features four studies, including (1): develop an image process technique to objectively evaluate and quantify sow's vulva size using a LiDAR camera; (2): develop a robotic imaging system to monitor individual sow's behavior for estrus detection using artificial intelligence; (3): develop an image processing pipeline to automatically evaluate sow's vulva size for estrus detection using the robotic imaging system and artificial intelligence; (4): develop a model to accurately identify the first onset of estrus after weaning using both behavioral and physiological (vulva swelling) traits as input. An image process method was first developed to extract a 3D vulva surface to quantify its size by manually selecting the region of interest in the point cloud collected using a LiDAR camera. Using this approach, vulva volumes of 8 sows, fed with synchronizer, were evaluated around their third post-weaning estrus. Results showed that no sow was found to have estrus before a significant increase in vulva volume was observed. This suggests vulva volume can be a reliable indicator of returned estrus. A robotic imaging system and a neural network model were developed to monitor changes in sows' posture (behavior) pattern prior to the first post-weaning estrus. The resulting posture recognition model can effectively classify sow's posture with over 98.6 percent test accuracy. The study also found that sows would have more physical activity and spend less time sleeping right before the first post-weaning estrus. In addition, an image processing pipeline was developed to automatically evaluate sow's vulva volume from the imagery data collected by the robotic imaging system through the use of multiple neural network models. The result showed the pipeline was able to extract the vulva region from the image data with a success rate of 96.7 percent. Furthermore, a noticeable increase in vulva volume was observed for all sows 0-1 day prior to their first post-weaning estrus. Finally, both behavior (posture records) and biological (vulva volume records) traits were used to help identify the first post-weaning estrus in sows. The trained estrus detection model showed a training accuracy of 95.4 percent, training specificity of 97 percent, test accuracy of 93.1 percent, and test specificity of 91 percent. In summary, this research demonstrated the success of robotic and artificial intelligence techniques in assessing important traits to help detect the onset of estrus in sows. The accurate and automated estrus detection in sows can help improve the efficiency, productivity, and sustainability of the pork industry while reducing the burden on swine farmers. Moreover, by detecting vulva swelling around the return estrus (about 21 days after artificial insemination), swine farmers can timely identify the non-pregnant sows. This can help reduce the non-productive days in a swine breeding herd and further reduce the labor needed for conducting pregnancy tests. Therefore, it is concluded that the developed robotic imaging system has great potential in improving swine production efficiency by timely identifying the onset of estrus in sows. This research could be scaled up to commercial swine breeding production and offered a paradigm for improving swine production and management efficiency.
Table of Contents
DOI
PubMed ID
Degree
Ph. D.