Skip to main content

DigiFarm revolutionises agriculture with artificial intelligence

The Norwegian company DigiFarm is on the verge of revolutionising agriculture with artificial intelligence (AI). Using advanced technology and access to European and national supercomputers, it is developing solutions that could have a significant societal impact. This article has previously been published on the Sigma2 website

This is particularly important for better utilisation of agricultural land, understanding human-made interventions in nature, and preventing deforestation.

DigiFarm’s precise field boundary data is a valuable tool for farmers and agricultural authorities worldwide. It helps them increase crop yields, adjust fertilisation and spraying, and create reliable crop forecasts. Additionally, the data can be used to assess environmental damage and monitor deforestation. In a time of raw material shortage, it is crucial that solutions quickly reach the market, especially given the challenges countries like Ukraine face with grain export.

Supercomputers, the key to massive product improvement

Through they research and development, DigiFarm has developed an AI model that can automatically detect field boundaries between fields. They train a deep neural network to identify boundaries and different elements in a field, such as grain, water, and trees. With 4 million hectares of training data from 57 countries, their model has become very large and requires significant computational power, or computational capacity, for training.

The model has recently been tested on LUMI.  LUMI is one of the world’s leading platforms in AI and is equipped with some of the highest-performing data processors (GPUs) available on the market. Supercomputers like LUMI use highly advanced and expensive GPUs, which are far more powerful than those in home PCs. These GPUs provide dramatically faster calculations, essential for Machine Learning. The training process is significantly accelerated through access to large amounts of interconnected GPUs, and bigger models can be trained while saving time. The results DigiFarm has achieved on LUMI are extraordinary. It will shorten the path from testing to a market-ready product by about 6 months.

—LUMI represents a turning point for us and will undoubtedly greatly impact our ability to deliver innovative solutions for the agricultural sector. Access to LUMI allows us to accelerate development and significantly improve our model’s precision, says Nils Helset, founder and CEO of DigiFarm.

—We have improved the model`s accuracy by 4.2% in just a few months. This is a major achievement in deep learning, as it becomes more difficult to achieve higher accuracy the longer a model is trained, says Helset.

Nils Helset, Founder and CEO of DigiFarm. DigiFarm uses artificial intelligence (AI) and supercomputers to streamline agriculture and monitor environmental damage. Photo: Anki N. Groethe.

Nils Helset, Founder and CEO of DigiFarm. DigiFarm uses artificial intelligence (AI) and supercomputers to streamline agriculture and monitor environmental damage. Photo: Anki N. Groethe.

4 million hectares of training data

DigiFarm has trained its model for five years before starting on LUMI, in commercial cloud environments and on the national supercomputers in Norway. They improve the resolution of images from Sentinel-2 satellites from 10 metres to 1 metre, using AI. With a comprehensive data foundation, the model learns to identify field boundaries and elements in a field, such as grain, water, and trees, and can now also detect the type of crop being grown.

In just 3.5 years, they have collected 4 million hectares of training data. Increased data volume makes the model complex and time-consuming to train. With access to many interconnected computational resources in the form of GPUs, they can accelerate the training process and significantly reduce the time it takes to develop a finished product. In just 2 months, DigiFarm has used a total of 104,000 GPU hours on LUMI for research and development.

DigiFarm improved the IoU accuracy of the model by 0.42. Intersection over Union (IoU) is a method that measures how accurately an artificial intelligence can identify and place objects in an image. This is a significant achievement in the field of deep learning, as it becomes increasingly difficult to achieve greater accuracy the longer a model is trained.

Field boundaries in Hamar Norway and in Central Italy.

Field boundaries in Hamar Norway and in Central Italy. Fields do not look the same in Hamar and Italy. The AI model must learn to recognise fields all over the world. Illustration: DigiFarm.

Untapped growth opportunities for companies

Access to national and European high-performance computing resources can be crucial for Norwegian companies wishing to conduct research and development to be competitive in a global market. Previously, it has mainly been researchers within academia who have driven the demand for such resources. Through Norwegian participation in the Digital Europe programme, this type of infrastructure is also made available to private and public actors. Now, demand is increasing among researchers in both academic and commercial environments.

—We must ensure Norwegian companies are on board when the train leaves the platform. This is especially relevant for Artificial Intelligence, where access to computational power is crucial for innovation and competitiveness, says Roger Kvam, head of the National Competence Centre for HPC, which has helped DigiFarm improve methods and utilise the national supercomputing (HPC) resources.

—DigiFarm perfectly illustrates how access to national supercomputers and Norway’s share of LUMI can accelerate technological development and enable new products. Our role is to support companies like DigiFarm in optimally utilising these resources, says Kvam.