AI Can Watch Mithun 24/7, Can It Help Farmers Spot Problems Earlier?

A farmer cannot watch every animal in a herd all day and night. Yet small changes in feeding, movement or resting can sometimes provide early clues about an animal’s health or reproductive condition.

Researchers at the ICAR-National Research Centre on Mithun in Nagaland have developed an artificial intelligence system that can watch Mithun continuously through cameras. The system can identify and track four behaviours: feeding, standing, lying and mounting.

The research raises a wider question for livestock farmers: Can cameras and AI become another pair of eyes in the shed?

How does the system watch animals?

The researchers installed 12 high-definition CCTV cameras across two sheds at the ICAR-NRC on Mithun farm in Nagaland. The cameras provide continuous day-and-night coverage, including infrared footage for monitoring when there is little light.

The team created a dataset of 3,000 manually labelled images showing four behaviours: feeding, standing, lying and mounting. The AI system was then trained to recognise these behaviours and follow individual animals across video frames.

The system uses YOLOv8n for behaviour detection and DeepSORT for tracking individual animals. In simple terms, one part of the system identifies what an animal is doing, while another follows that animal as it moves through the camera’s view.

The detection model recorded 99.5% mean average precision at mAP@0.5 and 99.6% recall in the researchers’ testing. It processed footage at about 31 frames per second using an NVIDIA RTX 3060.

Animals cannot tell farmers directly when something is wrong. Their behaviour can provide useful clues. Changes in feeding, standing and lying may indicate differences in comfort, nutrition or physiological condition. A farmer who knows an animal’s normal behaviour may notice when that pattern changes.

Mounting behaviour is useful for another reason. It can provide information related to reproductive and oestrus management, which can help farmers monitor breeding activity.

Today, much of this monitoring depends on people watching animals. That takes time and becomes difficult when animals need to be observed throughout the night.

An automated system could continuously record behaviour and flag changes for a farmer or livestock manager to examine.

Could AI detect illness earlier?

That is one of the possibilities researchers are exploring, but the current system does not yet diagnose disease.

The present study focuses on four behaviours. Researchers want to expand the system to recognise additional behaviours, including aggression, grooming and disease-related inactivity.

If future versions can reliably identify unusual inactivity or changes in feeding behaviour, they could potentially alert farmers that an animal needs closer attention.

The important point is that an alert would not be the same as a diagnosis. A farmer or veterinarian would still need to examine the animal and determine what is causing the change.

This distinction matters because the same behavioural change can have different causes. An animal may eat less because of illness, heat, changes in feed or other conditions.

Livestock breeding is another area where continuous monitoring could be useful. Farmers currently depend on observation to identify reproductive behaviour. Missing the right time can make breeding management more difficult. The Mithun system’s ability to detect mounting behaviour provides researchers with a way to collect more continuous information about reproductive activity. This could eventually support better monitoring of oestrus and breeding decisions.

The technology could become more useful if it can identify several behaviours together and track them over longer periods. For example, instead of looking at one isolated event, an AI system could potentially identify a change in an animal’s normal behaviour pattern over several hours or days.

Can this work on every farm?

Not yet. The researchers have so far tested the system at a single farm. It still needs testing across different farms, regions, seasons, animal densities and camera arrangements. Partial obstruction of animals can also affect detection and tracking. The current system also recognises only four behaviours. More data will be needed before it can reliably handle a wider range of animal activities and farm conditions.

This is important for small farmers. A system developed in a research farm may not automatically work in a crowded cattle shed, an open grazing system or a farm with different lighting conditions. Researchers are therefore exploring larger datasets, more behaviours, temporal AI models and the possibility of running such systems on edge devices closer to where the cameras are installed.

What could this mean for farmers?

The biggest benefit would be continuous monitoring without requiring a person to watch the herd every minute. A future system could potentially send a simple alert when an animal’s behaviour changes significantly. A farmer could then check that particular animal instead of inspecting the entire herd repeatedly.

That could save time and help farmers notice changes earlier. But the technology would need to be affordable, reliable and simple enough for ordinary livestock farms. For small farmers, cameras and computing equipment can also add costs. The value would depend on whether the system prevents losses, improves breeding or saves enough labour to justify those costs.

The technology should therefore be seen as a support tool, not a replacement for farmers or veterinarians.

The camera could become another pair of eyes

The Mithun study shows how livestock management is moving from occasional observation towards continuous collection of animal behaviour data.

For generations, farmers have learned to recognise when an animal is feeding normally, resting too much or behaving differently. AI is now being trained to recognise some of those same patterns.

The technology is still being tested and cannot yet promise earlier disease detection or better breeding on every farm.

But its direction is clear. The future livestock farmer may have more than experience and observation to rely on. A camera could keep watching the herd, software could track behavioural changes, and the farmer could decide when an animal needs attention.

The machine would not replace the farmer’s eye. It could give that eye a little more reach.