Automated berry harvesting solution for safer working conditions in the Finnish forests

Wild berries are one of Finland’s national treasures, growing naturally in the forests. Every year, in the short period between mid-July and late September, workers have to manually harvest approximately 500 million kg of berries. Coupled with the scarcity of berry pickers, long working hours and unsafe operating conditions have almost become the norm. Even more so since berry trading companies have invited pickers from Thailand to help with the harvest. While this initiative might seem straightforward, it comes with safety concerns. Not knowing the Finnish language, culture, and forests, Thai pickers can easily get lost or, even worse, injured.

Sparkd AI is delighted to partner with the FEROX consortium in devising an automated berry harvesting solution to create safer working conditions for berry pickers in the Finnish forests.

Lost in a Finnish forest

Imagine finding yourself in the middle of a thick forest with the task of picking berries but no map telling you where to go. You turn around, and all you can see are identical trees. You can’t find your bearings and don’t know what to do. Wouldn’t you be afraid of getting lost or injured and not being able to find your way out?

Although a little bit overdramatic, this scenario is not far from the conditions in which Thai berry pickers find themselves at times. Berries grow randomly in the wild, and recommendations on the best picking locations are based on assumptions, weather observations, and past years’ experience. On top of that, berry trading companies keep these locations secret to prevent the leak of information to competitors.

Not being able to speak the language and with little or no knowledge of the place, many foreign workers experience severe anxiety due to the fear of getting injured or lost in the forests. So, why do Thai people decide to travel to Finland to work in such precarious conditions?

Finnish forest
Picking berries in a Finnish forest

It’s all about the money. In 2021, the best berry pickers earned in two months the same money they would have earned in 15 years of work in Thailand. Earnings depend on the amount, the type, and the quality of the collected berries. For this reason, pickers often work long hours without breaks, adding extra stress to the already unsafe working conditions.

FEROX: a fully automated berry harvesting solution for safer working conditions

The FEROX consortium will design an automated berry harvesting solution to create safer working conditions for pickers. Ferox will deploy autonomous drones equipped with laser scanners, RGB cameras, and IoT sensors to collect data. This data will allow us to build models of the forests and make an accurate estimation of the berries’ locations, amounts, and types. Thus, we will develop specific AI models to help workers locate berries and optimise their operations. Moreover, FEROX will provide wild berry pickers with light-duty drones to help them find their way into the forest and heavy-lift drones to help them carry the harvest.

We ultimately want to prevent workers from experiencing anxiety or getting injured or lost while alone in the forest and to reduce the long working hours. For this purpose, the Ferox consortium will devise and implement tools that pickers can trust and rely on. In addition, we will ensure the workers’ safety through automatic monitoring and timely emergency aid when needed.

Sparkd’s role as lead Machine Learning and Computer Vision partner

Our team is very excited to be the lead AI partner of the FEROX consortium. Machine Learning and Computer Vision play a massive role in this project, and we look forward to bringing these solutions to life. Some of the CV and ML applications include: 

  • Detection and classification of vegetation types using above-canopy images. We will train deep learning segmentation models to detect the type of vegetation conducive to the growth of berries and mushrooms.
  • Fine-grained detection of berries and mushrooms from point clouds, images and video sequences of under-canopy drones and body cameras. We will train and test classification models to recognise individual berries with high commercial interest (e.g. bilberry or lingonberry/cowberry).
  • Multimodal and multisensor prediction of locations with a high likelihood of yield.
  • Development of an algorithm for fine-grained localisation of workers using Computer Vision. 
  • Develop an algorithm to suggest the next-best action in real-time if an incident event is triggered. The algorithm will consider all the available signals, including position, weather conditions, and position of other workers.

The project will kick off in September 2022, and we look forward to starting the work on this new challenge. The consortium estimates that the final solution will take five years to be implemented and properly tested. Until then, keep visiting our blog for updates on this interesting project.

Author Manuela Armini & Giacomo Piccinini

Photo by K8 on Unsplash

Start your Ai journey today!

Gemmo's noise classification case study with Sonitus