Whether for milking, feeding or health monitoring – AI applications are ready for practical use in many areas of dairy farming. However, they are only truly useful if farms make full use of all the data collected.
Systems based on artificial intelligence (AI) are already widely used on dairy farms. In some cases, however, they are ‘hidden’ and not explicitly labelled as AI, as the functions are often directly integrated into standard technology for milking, herd health or feeding. But how does AI differ from sensors? Whilst sensors measure but do not ‘know’ what the data means, AI can continuously analyse large volumes of sensor, camera and management data. It recognises the patterns underlying this sensor data. As an analysis tool, AI generates alerts, assessments and recommendations from the raw data. However, it cannot predict the future; it can only draw conclusions from large volumes of data.
Health monitoring in the barn
AI is already well established in health monitoring. This applies to the early detection of diseases or behavioural changes, enabling early diagnosis and appropriate treatment. Hanna Strodthoff-Schneider, Managing Director of the veterinary practice agro prax, says: “In our work with dairy farms, we use various data provided to us by the farms. This includes sensor data, data from milking robots, as well as analyses from the MLP and herd management programmes. Sensor data, for example, enables us to identify not only individual sick animals but also herd-wide changes resulting from alterations in management or feeding practices at an early stage.’
As a rule, the farmer grants the vet access to their data in the herd management programme for this purpose. “For our AI-based analyses, it would be helpful if there were more interfaces between the sensors. This is still a major problem at present. Due to a lack of interfaces, for example, it is often not possible to view all the data at a glance. This costs time and there is a lot of untapped potential here,” says Hanna Strodthoff-Schneider. “Ketosis is one example. Of course, even a sensor cannot detect every instance of developing ketosis, as every cow reacts differently to the condition. And in some cases, it does not even require treatment.”
The situation is different if ketosis also leads to a drop in milk yield. In such cases, it would be useful not only to have data on feeding and rumination behaviour, but also direct measurements of milk yield.”
Detecting diseases earlier
“Activity monitoring reliably identifies sick cows, even at what are sometimes very early stages. This significantly reduces the workload involved in locating the animals, which in turn means shorter time spent standing for them. I often look at a cow’s behaviour pattern on a graph and can already tell, for example, the difference between an animal suffering from milk fever and one with mastitis or metritis,’ says Hanna Strodthoff-Schneider.
“I am quite certain that, from a technical point of view, it will be possible in the coming years to make preliminary diagnoses using AI-supported systems. The combination of sensor technology and clinical examination could be used as an opportunity to improve diagnostic accuracy even further. “Early diagnoses reduce the treatment burden, lower the use of medication, minimise the drop in milk yield and cut down on the workload. Image and speech analysis is one of the trends. ‘This involves, for example, systems for the automatic detection of lameness in cows,’ explains Maria Schneider of the Bavarian State Research Centre for Animal Production (LfL Bayern) and project coordinator for the DigiMilchPro pilot project. The aim here is to use computer vision to analyse deviations in movement and behaviour patterns, as well as climate, behavioural and performance data, in order to automatically detect lameness. Voice and app interfaces that support herd managers using natural language are already market-ready for calves, but are also set to become relevant in dairy cattle barns in the future.”
The use of AI in feed production
AI is already playing a role in grassland management and silage production too. “We use it, amongst other things, to estimate yields,” says Maria Schneider, “for example, models are trained using weather and soil data, as well as vegetation indices derived from remote sensing data. They can then estimate grassland yields across the board. This already works reasonably reliably for biomass growth; it is only for dry matter yields that the models are still too imprecise (in Germany; in Austria, there is a model that already works well).“
Conclusion: AI can be used in many ways
“The sensor technologies work reliably on dairy farms. However, too little use is currently being made of the data obtained from them,” says Maria Schneider. This is where research and industry come in: “The complex sensor data must be processed using AI approaches in such a way that farmers can – and want to – make full use of it without too much effort,” says Maria Schneider. Key to this is not only the farmer’s confident use of digital technologies but also an understanding of their possibilities and limitations. People should check and question the AI results and make decisions themselves.
Using AI to organise the farm office better
In a farm office, AI is particularly useful for all tasks that are repetitive, text-heavy or organisational in nature. “For AI to be used effectively there, the farm’s entire process chain must be digitised,” says Dennis Welleweerd. He is the managing director of the consultancy Farmers Factory and also runs a dairy farm.
AI cannot process printed invoices. These must therefore first be scanned and thus digitised. “Invoices, documents, etc. should be stored in a single digital location, e.g. in the cloud, which the AI can access.
A recurring hurdle when using AI to link sensor data is the lack of interfaces, for example with the herd management programme or the field register,” says Dennis Welleweerd. The AI must be set up in such a way that the farm’s individual tools work together seamlessly. To achieve this, most farmers commission an external service provider.
Team meetings, reports and task lists
‘For example, if I record a conversation, the AI transcribes the recording into text. In the next step, I can have summaries generated from this, or I can have the AI assign tasks to staff,” says Dennis Welleweerd. There is also growing demand amongst his clients for farm-specific apps, for example for time tracking, where off-the-shelf models are good but need to be adapted slightly to the specific requirements of the farm.
‘Onboarding apps’ are becoming increasingly popular
Onboarding apps help new employees or trainees familiarise themselves with company processes, tasks and specific requirements. With checklists, explanatory videos and work instructions accessible via the app, they can be trained more easily and in less time than if the farm manager had to show them everything in person. This saves time for the employer and makes it easier for new recruits to settle into the job. The app is mainly planned by the farmer and the consultant. The AI then builds it according to their specifications. The farmer can then provide the AI with instructions for corrections and additions, which the AI in turn incorporates into the system independently. The development of AI tools for the farm office is progressing rapidly. Dennis Welleweerd is convinced that we are still in the early stages of AI adoption in agriculture and that the possibilities will grow enormously.
Bianca Fuchs
