Automated 3D Imaging and AI-Driven Monitoring are Redefining Reproductive Management, Health Assessment, and Labor Efficiency in Modern Sow Barns

By Dr. Ziteng Xu

Precision livestock farming (PLF) is rapidly transforming how producers manage breeding herds.

While early PLF tools focused primarily on environmental monitoring and group-level data, the next step in precision sow management is automated, individual-animal monitoring that supports reproductive decisions and health assessment in real time.

To achieve this goal, researchers from the University of Missouri and Texas A&M University (Smart Swine Team) are developing a robotic imaging system (Figure 1) that captures 3D images of sows at 10-minute intervals.

The system is designed to perform automated estrus detection, early pregnancy diagnosis, and overall health monitoring using non-contact 3D cameras. This multifunctional tool aims to improve reproductive efficiency, reduce labor demands, and increase sow longevity by providing continuous insight into each sow.

WHY PRECISION SOW MANAGEMENT MATTERS

Reproductive performance drives profitability in breeding herds. Missed estrus events, delayed breeding, undetected pregnancy failures, and health issues all contribute to non-productive days and increased culling rates.

Traditional monitoring relies heavily on manual observation, including back-pressure testing, visual inspection, and technician experience. These methods are labor-intensive and can vary in accuracy depending on staffing levels and skill.

Precision sow management shifts this model from having an occasional manual observation to frequent automated monitoring.

By collecting data throughout the day, PLF systems can detect subtle changes in behavior, body shape, and physiological indicators that are difficult to observe during routine barn checks.

This allows producers to make timely decisions and focus attention on animals that require intervention.

AUTOMATED ESTRUS DETECTION

Maintaining accurate estrus detection is critical for breeding sows at the correct time and maximizing reproductive performance.

Advances in artificial intelligence and imaging sensors now allow behavioral and physiological signs associated with estrus to be measured automatically.

These indicators include increased activity level, changes in posture, vulva swelling, and more. The robotic imaging system captures images of each sow throughout the day and extracts these features automatically (Figure 2).

Artificial intelligence models then analyze changes over time to detect the onset of estrus and generate alerts for breeding. Continuous monitoring ensures that estrus events are identified even when staff are not present, improving consistency and reducing the risk of missed heats.

In a pilot trial involving 60 multiparous sows (Figure 3), the developed estrus detection model achieved 93.3% testing accuracy in detecting the onset of estrus without false alarms.

These sows were confirmed with positive pregnancy status and farrowed successfully.

Ongoing work aims to further improve accuracy to approximately 98% by increasing sample size and incorporating additional features such as perked ear duration during boar contact. This level of accuracy has the potential to significantly improve both breeding timing and overall reproductive performance.

Automated estrus detection also improves labor efficiency. Instead of manually checking every sow multiple times per day, staff can focus on animals flagged by the system. This targeted approach reduces labor demands while maintaining or improving reproductive outcomes.

EARLY PREGNANCY DIAGNOSIS

Timely confirmation of pregnancy status is essential for minimizing non-productive days.

Traditional ultrasonography is typically performed several weeks after breeding, which can delay identification of open sows and postpone rebreeding. Precision sow management systems offer a complementary approach by detecting returned estrus, which indicates failure to conceive.

Monitoring physiological indicators allows the system to identify sows that return to estrus approximately three weeks after breeding. In preliminary observations, behavioral signs of returned estrus were relatively subtle compared to the initial post-weaning estrus (Figure 4).

However, vulva swelling remained a consistent indicator.

Using vulva swelling alone, the system detected first post-weaning estrus with approximately 95% testing accuracy, with false alerts occurring only one day before the actual onset of estrus for six of the sows involved.

During the pilot trial, one of the sows had a false negative pregnancy test result from the ultrasonography approach approximately 35 days post-weaning. The false-negative diagnosis was overturned by the algorithm as there was no pronounced vulva swelling detected around the expected return of estrus (Figure 3).

This approach enables earlier identification of non-pregnant sows and allows timely re-service, reducing non-productive days. Because this method relies primarily on physiological indicators rather than activity tracking, it may also be applicable in group-housed systems where individual behavior tracking can be challenging.

This non-contact pregnancy assessment approach also reduces labor requirements. Instead of scanning all animals manually, producers can prioritize sows flagged by the system. This improves efficiency while maintaining accuracy in pregnancy monitoring.

OVERALL HEALTH ASSESSMENT

Beyond reproduction, precision sow management systems provide continuous health monitoring. Changes in behavior, posture, or body condition often occur before clinical signs appear.

Detecting these early indicators can help prevent production losses and improve animal welfare. Algorithms are under development to enable several health-related measurements, including:

• Body Weight and Body Condition: 3D imaging allows estimation of the sow’s body volume and heart girth length to track weight changes over time (Figure 5a). This helps to better identify excessive weight loss, poor body condition, or feeding issues. Continuous monitoring also supports management decisions related to gilt development, breeding readiness, and sow longevity.
• Structural Correctness and Mobility: Computer vision algorithms can evaluate activity, posture, stance, and 3D leg structural key points (Figure 5b). These measurements help identify lameness or structural issues that may reduce productivity or increase culling risk. Early detection allows timely intervention and improved welfare.
• Respiratory Rate Monitoring: The system also estimates respiratory rate by tracking flank movement (Figure 5c). Elevated respiration can indicate heat stress, respiratory disease, or environmental challenges. Continuous monitoring allows producers to identify at-risk animals and evaluate cooling strategies more effectively.
• Activity and Behavioral Monitoring: Activity patterns, posture changes, and resting behavior provide valuable insight into health status. Reduced activity may signal illness or lameness, while increased restlessness may indicate discomfort or stress. Tracking these patterns continuously improves early detection of health issues.

While each indicator provides useful information, relying on a single metric can be misleading.

For example, reduced activity alone may indicate illness, but it could also reflect normal resting behavior. Similarly, elevated respiration may be caused by heat stress, excitement, or disease.

Multi-metric inspection integrates several indicators—such as activity, respiration, body condition, and posture—to build a more complete and reliably consistent assessment of sow health.

When multiple indicators change simultaneously, confidence in detecting the root cause of the health problem can increase significantly, which provides farmers with reliable, actionable alerts in a timely manner.

CONCLUSION

As labor availability continues to tighten and production efficiency becomes increasingly important, technologies that provide automated, data-driven decision support will play a growing and increasingly vital role in breeding herd management. Multifunctional precision livestock farming systems represent a practical next step—offering producers a single platform to monitor reproduction, assess health, and manage sows more efficiently.

Dr. Ziteng Xu
Texas A&M

Dr. Ziteng Xu is an Assistant Professor at Texas A&M AgriLife Research. He is also a co-founder of Smart Swine, LLC. His research program focuses on developing precision livestock farming technologies and sustainable livestock production systems.