This company uses machine learning to tell cows apart from each other2 min read . Updated: 13 Sep 2019, 08:25 PM IST
- Like human faces, cows and buffaloes also have specific features on their faces that can be recognised
- The technology can be used to deal with the problem of stray cattle in a country where lack of shelter and food leads to more than a crore stray cattle
Have you ever been able to tell one cow apart from another? Well, your phone can. Like human faces, cows and buffaloes also have specific features on their faces that can be recognised, says Aashna Singh, co-founder of a startup called MoooFarm. The company has developed a machine learning-based solution that can tell cows apart and help deal with the problem of stray cattle.
“If you can unlock your phone with your face, why can’t it recognise cattle?" Singh asks. That’s the question the company set out to solve when they first came up with the solution. MoooFarm’s business actually revolves around developing technology solutions for farmers, helping them manage their cattle better.
According to Singh, MoooFarm’s algorithm was developed by taking numerous photos of cows’ faces. The company took photos from different angles, in different lighting and even in different backgrounds. For instance, looking at a cow’s photo with different backgrounds teaches the ML model that it won’t always have a barn wall as the background.
Similarly, the different photos also help it distinguish between different breeds, ages and so on. The algorithm takes 300 pixels from the photo of a cow’s face. “They may look the same (to us), but for the algorithm they have absolutely different features," Singh said. She suggested that while two cows may look the same, there are distinguishing features in their eyes and the muzzle, which the algorithm uses to tell them apart.
Imagine the dot projector on Apple’s iPhones. The philosophy here is the same, but it doesn’t use infrared technology, instead depending on a basic smartphone camera. Taking photos of the cattle’s face from different angles serves the same purpose that rotating your face in front of the iPhone’s camera does when you’re training the FaceID algorithm to identify your face.
The company is currently working on identifying what the minimum resolution or quality of a smartphone needs to be where its algorithm works. Singh did say though that they have been testing with “basic" smartphones.
MoooFarm claims its facial recognition model is 95.7% accurate at the moment and the company is refining it further right now. The facial recognition technology won a $30,000 award from the World Bank in the Data Analytics Category.
But why do we need facial recognition for cows? According to Singh, the fact that cattle look similar opens opportunities for insurance fraud. The company is working with insurance firms to use its technology for telling cattle apart. Instead of existing solutions like RFID or placing tags on a cow’s ear, it can use facial recognition to tell whether the cow in question is insured or not.
Further, you could also use the technology to deal with the problem of stray cattle. According to reports, lack of shelter and food leads to more than a crore stray cattle in the country. Allowing an app to recognise these cattle could help keep them in check. MoooFarm’s solution also helps create a directory for cattle, noting exactly which farmer a cow/buffalo belongs to and other data about it.
MoooFarm isn’t the first company to try facial recognition in cows either. An Irish-startup called Cainthus uses imaging to tell animals apart from each other, using overhead cameras and drones. Researchers at the University of Kentucky had also been working on utilising drone cameras to identify type of cattle. India though, is primarily comprised of farmers with small holdings, making it more difficult to use drones and overhead cameras. Smartphones allow a more personal, accessible and individual solution.