Skip to main content

Grow tracking using machine learning and camera visualization vehicle



Problem

A rapidly growing company in the vertical farming industry that grew variety of crops, including leafy greens, herbs, and small plants, was challenged by increasing scale and complexity of their operations. Their existing manual inspection methods, to verify that plants were growing as they should, were becoming less efficient and more time-consuming. The method was prone to errors and required highly skilled employees to do repetitive tasks. To uphold the quality of the products, prepare for scaleup, catch errors, reduce overproduction and loss, a new more automated method had to be found.

Solution

To address this challenge, the company contacted Nubis to implement a vision-based grow tracking system, utilizing machine learning algorithms to detect abnormalities, identify crop age, seeding errors, system errors and pests.

After some study, the solution was scoped to solve three areas:

  1. Collect image data on a regular basis in an efficient manner without too much hardware cost.
  2. Implement and train AI algorithms to detect abnormalities based on thousands of pictures of the subject.
  3. Create dashboards, graphs and Cloud software for the operator.

Nubis decided to build a semi-autonomous vehicle with a series of cameras that automatically recorded videos of all crops from different angles, allowing it to cover the whole farm, and to save greatly on number of cameras and equipment. The video feed was uploaded to the Cloud, broken down to identify each set of plants using aruco markers and fed through machine learning algorithms to identify among others: The type of plants growing for verifying with the planning software; give age prediction to verify with the seeding date; look for abnormalities to verify for pests.

The data was collected to a Cloud platform and correlated with the farm control and planning software and presented for the operator through alarms, graphs and actual pictures. The installation required no disturbance to the farm operation and minimal addition of new hardware.

Conclusion

By partnering with Nubis, and after the system was installed, the company was able to detect over 98% of the problems and instead of having qualified employees walking around the farm, they would automatically receive notifications about farm issues which they could asses directly from their office chair.