Singapore seems an odd place for a start-up that wants to help pig farmers. The country began phasing out pig farming in 1984 and closed the last farms in 1989. But that has not stopped Howard Tang, CEO and co-founder of Smart Animal Husbandry Care (SmartAHC), from building his company here. “It sounds ridiculous that there is not a single live pig in Singapore,” says Tang. “But what we want to do here is to become the e-enabler for smart farming rather than do the farming itself in Singapore.”

SmartAHC has developed a wireless sensor that monitors a sow’s temperature and activity. The sensor feeds the data into software that uses unsupervised machine learning to determine a sow’s estrous status — whether it is in heat. Today, this process of determining if a sow is in heat is tricky and has varying success rates across farms. Farmhands use tests such as a back press test, in which the sow arches its back when weight is applied at its loin. The more experienced the farmhand, the more accurate the result of the test.

“The outcome is very subjective, depending on the experience of the farmers,” says Tang. “What we normally see is a 75% success rate in sows sent for artificial insemination. And the number fluctuates in big farms, as farms in many cities are managed by workers with different [levels of] experience.”

The temperature of a sow will increase by 0.5˚C when in heat. But temperature increases can also indicate that a sow is sick. Temperatures also fluctuate based on activity levels. SmartAHC therefore uses artificial intelligence to look for patterns and make accurate predictions. “Previously we made use of artificial neural networks, a subset of artificial intelligence. But recently, we have moved on to unsupervised machine learning, which is more sophisticated and provides higher accuracy,” says Tang. “Machine learning is good at pattern recognition.”

The software is also able to produce a dynamic model and make allowances for biological differences between pigs, Tang explains. The software will keep learning through pattern recognition, picking up indicators during the estrous cycle. Currently, the software has been tested with up to 60 sows at once. The aim is to be able to track 1,000 pigs before a commercial launch.

“The good thing is that most of the pigs in China are Yorkshire pigs, so a model built for that breed can be used for most of the farms,” says Tang. “Pigs still react differently in the same environment and if they are of the same breed. Machine learning is able to, through indicators, make decisions based on experience.”

SmartAHC is also developing a product that will help medium- to SmartAHC uses wireless sensors to help farmers monitor pigs for better breeding large-scale farm owners to remotely manage their workers. The product will track the health of workers. “We realise that there is a shortage of farmers worldwide and we want to incorporate the experience of a farmer into a computer system to improve the quality and productivity of workers,” says Tang.

Testing challenges
The stories Tang tells of SmartAHC’s early days would be familiar to any entrepreneur. The earliest designs of the wireless sensors were inserted into the heads of dead pigs. The team would then shake the heads as hard as possible to see if the sensors dropped out. “But when we sent it over to China [to test on a live pig], the sensor dropped out after two seconds,” says Tang.

SmartAHC had teamed up with Sichuan Agricultural University in China and asked students for help to test out its designs. But the students had trouble communicating to the designer in Singapore as to why the sensor dropped off, Tang says. The company produced up to three iterations with no improvement.

“That’s when we decided to set up a small office in Shanghai, China, and keep four pigs in a small farm about 20 minutes’ drive from our office,” says Tang. A mechanical engineer was also recruited in China to eyeball design issues firsthand and rectify them. This resulted in design times improving tremendously. The time it took to make significant improvements went down from one year to as little as two to three months. With the new office, prototypes could be made in China and tested directly on live pigs. “Now, when our design is done in China, we can get the prototypes within three days and test it on live pigs, bringing testing time down to five days,” says Tang.

SmartAHC has now progressed to commercial trials with a farm in China, having achieved 95% accuracy in a controlled test environment. Tang admits that the first phase of commercial tests on electronic and mechanical stability failed due to issues such as waterproofing and the wire being torn off by the pigs. “But through this, we knew which area to strengthen,” he says.

The second version of the improved design for the sensor has seen better results, continuing to collect data after two weeks. The next phase of tests will be on the algorithm to determine the estrous status of a sow.

“Starting early December, we will test on breeding sows ready for mating. We will test our algorithm to see if we can achieve the same accuracy (95%) achieved in a test facility in Sichuan Agricultural University,” he says. “The objective for the commercial test is to hit 80% to 85% accuracy.”

Targeting mega-farms
An alumnus of the Nanyang Technological University, Tang got the idea for a pig sensor when working on his PhD in electrical and electronic engineering. Tang and SmartAHC co-founder Lan Song, a fellow PhD candidate hailing from China, joined NTUitive, NTU’s start-up accelerator programme. Their idea received undisclosed seed funding from the Small World Group in 2015, and another $1 million from GreenMeadows Accelerator and the Spring Singapore SEEDS programme. SmartAHC is currently closing another funding round in China for RMB8.2 million ($1.7 million).

Four years on, SmartAHC has signed letters of intent with five farms in China and worked with three of them for commercial tests. The target is to launch its wireless sensor and software product by April next year, according to Tang, with plans to convert the five farms into paying customers. These five farms are mid-range farms with 10,000 pigs. Tang has ambitions to target mega-farms with 100,000 breeding sows by 2018.

Another product in the works will help authorities track whether pigs in farms have been tagged. Pigs slated for consumption have to be tagged by law. This is to prevent farms from dumping dead pigs into the river or selling them at a wet market. But many farmers only tag their pigs days before they enter the slaughterhouse, where visual inspections are done. “We call it Smart Ear Tag. It is able to detect whether a tag is attached to the pig or not, helping the government better enforce the law on tagging the pig,” says Tang.

There are also plans to go into new products such as weight management for pigs. And SmartAHC is talking to farms in Indonesia and Vietnam to bring the product to Southeast Asia and even the US after a successful launch in China.

“Longer term, [our goal] is to cross into other livestock,” says Tang. “We do receive a lot of requests from chicken farms, so we may build a system for chicken farms as well.”

This article appeared in the Enterprise of Issue 758 (Dec 12) of The Edge Singapore.