AI (Artificial Intelligence) and robotics have the potential to significantly boost the foraging yields of wild berries in Europe. Wild berries are a nutritious and delicious food that can offer a variety of health benefits. They are packed with antioxidants, vitamins, and minerals, and they have been linked to several positive health outcomes. 

Berry collecting is a deeply ingrained tradition in Scandinavian countries. Furthermore, the tradition holds both cultural and practical significance. Despite the abundance of berries, less than 2% of berries are harvested.  

The FEROX Challenge

The FEROX project is approaching the challenge of transforming berry collecting through the development of hi-tech solutions, that will attempt to identify and map berry patches. This will enable the optimisation of foraging routes when the berries are ripe. At the same time, the project aims to monitor wild berry populations and track changes in their distribution and abundance to ensure the long-term health of the forest.  

FEROX is developing AI algorithms to integrate satellite imagery, geospatial data and video feeds from under-canopy drones to identify areas with a high potential for wild berry growth. As well as this, the algorithms will be trained to observe berry flowers, unripe fruit and then ripe berries. This information will be used to create AI-powered foraging apps to provide foragers with real-time information about berry patches, ripeness, terrain conditions and weather conditions. Such information will guide foragers to the most productive berry patches, optimising their foraging routes. 

AI models are also being developed that can predict berry ripeness, based on berry growth patterns, prior observations and environmental factors. This is to reduce the overhead of data gathering missions in forests, i.e., the flying of drones, gathering visual data and then analysing results.  

Boosting Yield through AI & Robotics 

The FEROX project represents a pivotal advancement in AgriTech, primarily focusing on the development of advanced berry detection systems through the application of deep learning algorithms. The initial phase of this project involved an extensive data collection process, executed using cameras affixed to drones. These drones conducted aerial surveys of forested areas, capturing images crucial for the foundational training of our AI models.  

An integral component of data preparation for AI training was the involvement of human annotators. These experts engaged in the rigorous process of meticulously labelling each image captured by the drones. This process was vital to ensure the precision and accuracy of the data fed into the AI models. It resulted in detailed datasets encompassing diverse berry species like lingonberries, bilberries, crowberries, bog berries, and cloudberries. 

Equipped with this accurately annotated data, we proceeded to train our object detection algorithms. These algorithms are intricately designed to identify and classify various types of berries from the drone-captured images. The training phase of these models necessitated a delicate balancing act between achieving high accuracy and maintaining computational efficiency. 

The challenge inherent in this project lies in the differentiation of various berry types; a task that our preliminary results have shown to be feasible albeit with some limitations. Specifically, the algorithms have demonstrated a commendable degree of accuracy in recognising most types of berries. However, the distinction between certain berry varieties, such as bog berries and bilberries, remains a challenge due to their close resemblance in shape and colour. 

Despite these hurdles, the project team maintains a positive outlook. The initial successes in achieving notable accuracy in berry detection bolster our confidence in the project's future trajectory. Continuous refinement of the AI models and further enhancement of their learning capabilities are expected to yield improved performance, particularly in differentiating between more complex berry classes.