Yield Systems democratizes crop data production by using smartphones, low-cost hardware (HW) and high-end synthetic data-based video intelligence to estimate agricultural and biological properties of crop canopies.
Yield Systems’ proprietary crop data production system is based on spreading the canopy to obtain full visibility to all above-ground plant traits (unlike drones, that only see the top 10 cm of the canopy).
From high-resolution video, we count and estimate the same traits that a human agronomist/plant breeder would look at: number of spikes, seeds/spike, seed size, biomass, plant height, disease symptoms and spatial variation of the above traits. Importantly, the estimated traits are maximally interpretable to all farmers (unlike spectral intensities etc obtained with many competing technologies).
The device is built on a high-end smartphone. In two-three years time, regular smartphones will have the camera capabilities necessary for our video intelligence and our solution...
Yield Systems democratizes crop data production by using smartphones, low-cost hardware (HW) and high-end synthetic data-based video intelligence to estimate agricultural and biological properties of crop canopies.
Yield Systems’ proprietary crop data production system is based on spreading the canopy to obtain full visibility to all above-ground plant traits (unlike drones, that only see the top 10 cm of the canopy).
From high-resolution video, we count and estimate the same traits that a human agronomist/plant breeder would look at: number of spikes, seeds/spike, seed size, biomass, plant height, disease symptoms and spatial variation of the above traits. Importantly, the estimated traits are maximally interpretable to all farmers (unlike spectral intensities etc obtained with many competing technologies).
The device is built on a high-end smartphone. In two-three years time, regular smartphones will have the camera capabilities necessary for our video intelligence and our solution can be used to produce extremely accurate estimates of canopy properties globally, also in the developing world. We are also working on integrating this accurate data with satellite imagery to further increase the coverage of the data produced.

A central bottleneck for our solution is obtaining access to a sufficient amount of training data to train deep neural networks to process the wide range of traits needed for various crops and needs. For this, Yield Systems uses synthetic data, i.e. we train our neural networks in virtual reality to avoid the bottleneck of data availability. Synthetic data for training machine vision models is an emerging hot research topic; our team pioneered this by starting synthetic data -based training in 2017 while still a part of Aalto University. In other words, we use simulation technology to speed up the breeding process by simulating training data for machine vision algorithms with 3D models to eliminate the data bottleneck that otherwise limits the adoption of machine vision. Machine vision, on the other hand, is the key to cost-efficient data production and the improved data production capability further paves the way for adoption of other AI, whose capabilities depend on the quality of input data.
More information

Employees

Harri Juntunen
Admin
Harri Juntunen COO Generalist and entrepreneur with passion to make our world more sustainable by building innovative businesses.
Ilari Nieminen
Admin
Ilari Nieminen Engineer Developer