)
)
How can we help you?
)
Marija Mijic
HR & Office Allrounder
T +43 660 383 95 46
office@craftworks.at
craftworks GmbH
Schottenfeldgasse 20/6a
1070 Vienna
Yield Prediction with
Artificial Intelligence
Predict the yield of wheat grain fields through sample images on the ground, two weeks before harvest.
:quality(80))
Success Story
reading time: 3 min
Energy & Utilities
Predictive Quality
Predictive Maintenance
Visual Inspection
95%
Accuracy per square meters
1 min
needed per square meters
No damage
Or interference with the plants' growth, to the crop
Context
The Austrian Haill Insurance covers crop failures. Droaught is increasing the number of claims and due to internationalization, the company is facing severe competition. For claim management, fields need to be sampled two weeks before harvest, which is a very tight window of opportunity. All these factors make this task almost unmanageable for surveyors. Samples need to be taken by hand and brought into a laboratory. This procedure damages the plants.
Challenge
The ÖHV challenged craftworks and other companies to come up with a concept for swift and easy prediction of the yield of wheat fields at a point in time in the future. Particularly challenging was that not only the yield per plant varied, but also the yield per m² within a field varied greatly. Plants cluttered, were not distinguishable, and up to 40% of the plants were not ripe enough to count towards yield. craftworks won the challenge and was invited to build an MVP, which included hardware scouting.
Assignment
To predict the yield of a full field two weeks before harvesting, through geo-referenced pictures of samples taken by the surveyors on the ground, with a portable device, without needing to damage the plants for laboratory samples. A further goal of maximum 1 minute testing time needed per plant was set.
)
)
Solution
craftworks found a high spectral resolution LIDAR camera with which 1 m² of wheat field could be photographed through infrared. This enabled us to detect single wheat grains, measure their moisture content and extrapolate these findings in a model that we trained with the help of existing sample results from the laboratory. Results were very promising, however a much greater number of samples will be needed to improve its accuracy.
craftworks' AI system achieves 95% accuracy in predicting wheat yield per 1 m² using advanced imaging technology for moisture detection.
:quality(80))