Worlds First AI-Based Pregnancy Detection for Pigs

A new pathway has opened for non-experts to easily determine the pregnancy status of sows a task traditionally performed by skilled professionals.

The Rural Development Administration (RDA) (Commissioner Kwon Jae-han) announced the world’s first development of AI-Based Pregnancy Detection Technology for Pigs which allows non-experts to quickly and accurately determine the pregnancy status of sows using artificial intelligence (AI).

Typically the pregnancy status of pigs is confirmed by observing the sows behavior starting from 21 days after artificial insemination and reading uterine ultrasound images from 25 days onward when the fetal sac is clearly visible.

However the accuracy and timing of pregnancy detection using ultrasound images greatly depend on the skill of the operator and non-experts generally cannot determine pregnancy until 28 days after insemination.

The research team collected over 200000 high-resolution (5MHz) uterine ultrasound images and developed an AI model that can determine pregnancy status by applying a computational algorithm based on AI learning.

Additionally the team collected 230000 low-resolution (3MHz) uterine ultrasound images commonly used in pig farms and applied image enhancement technology to these images. They trained the AI model to create a farm-ready pregnancy detection program for pigs.

By capturing abdominal ultrasound images of sows for more than 10 seconds and running the program the AI determines and indicates pregnancy status. The technology demonstrated an accuracy of over 95% when used between 22 to 25 days after artificial insemination.

Accuracy by pregnancy days: (18-21 days) 88.4% (22-25 days) 95.7%
The RDA stated that this technology allows non-experts such as novice farmers and foreign workers to easily confirm the pregnancy status of sows. The process is quick and accurate improving operational efficiency for farms that often face challenges due to frequent staff changes or lack of skilled personnel.

Earlier pregnancy detection allows for extended health management periods for pregnant sows leading to increased productivity. Moreover non-pregnant sows can be re-inseminated sooner reducing non-productive days* and lowering feed costs.

Non-productive days: Total days a sow is neither pregnant nor nursing.
The RDA has filed three patent applications* related to the AI-Based Pregnancy Detection Technology for Pigs and plans to transfer the technology to interested companies. The RDA also aims to enhance the AI model to achieve over 95% accuracy for pregnancy detection from 18 to 21 days.

Patent applications: Training device for deep learning model for pig pregnancy diagnosis device for generating training data pregnancy diagnosis device and methods thereof (10-2024-0100652) among others.
Lee Chang-beon the owner of DDFarm in Yeoncheon Gyeonggi Province who participated in field testing said “Training non-expert workers like foreign employees to read uterine ultrasound images required a lot of time and effort. This technology allows easy and stable pregnancy detection providing significant help.”

Im Ki-soon Director of the National Institute of Animal Science at the RDA stated “We are also developing various AI technologies for managing sow body shape and detecting risks in piglets in addition to pregnancy detection. We will continue to expand smart farming technologies that can enhance productivity and reduce farming costs thereby strengthening the competitiveness of pig farms.”

Website: : http://www.wip-news.com/news/articleView.html?idxno=24556