Digitalization of Agriculture

Under the consortium management of the FZI Research Center for Information Technology, the project successfully brought together the expertise of all partners. The start-up SAM-DIMENSION GmbH, for instance, developed unmanned aerial systems (UAS) with unique sensor technology to create high-resolution maps. The project partner offers drone-based weed mapping that helps to recognize emerging weed problems and reduce herbicide application through more targeted use. In the next step, the FZI used the generated high-resolution maps to identify crops such as corn, winter wheat, and barley on a field and individual plant basis and analyzed them using image evaluation and AI methods. With the expertise of its inspectors, Landwirtschaftliches Technologiezentrum Augustenberg (LTZ) contributed domain knowledge, training data, and a manually collected baseline to the successful project conclusion: AI can now take over plant evaluation for automated assessment.

The evaluation was done in an integrated pipeline with minimal human intervention. From the high-resolution georeferenced UAS images, individual plants were first determined using a so-called Siamese neural network. In the next step, these plants were analyzed for selected evaluation parameters using a MultiNet approach. For the training of the AI, the extensive image data from various LTZ field trials and from state variety trials in corn and cereals (winter wheat, winter barley, and spelt) over three years were used and labeled manually by experts. The results show that the AI works with excellent reliability and recognizes relevant features even under challenging conditions.