Open image database for plant cultivation

Purpose and goal

The project aims to create an open large image database which through machine learning and computer vision can find where a certain crop grows in a field and its health & status. The goal is an image database of various named crops, both crops & weeds. This can then be used as a basis for decision support systems that help the farmer decide when it is time to control weeds. In the future, the information could be used for direct control of efforts, or to determine the proportion of different grass & clover varieties there are in a grassland, which gives how one should fertilize to keep the clover.

Expected effects and results

The collected data will be published for free use and thus strengthens basic research in automatic recognition of crops relevant to Swedish conditions and its diseases. It creates new opportunities for start-ups to develop innovations for this area without being burdened by costs for data collection. Preliminary analyses will provide indications of results that could be achieved in future applications. These can serve as a first input to environmental authorities about what is possible with new technology, and thus be the basis for any future legislation.

Implementation

The project will use testbed Digitalised agriculture which is run by RISE in Uppsala to develop both collection methodology and systems. It will continuously collect images during all growth phases from sowing to harvest using flying and ground-based camera systems. The use of two different carriers of collection systems is justified by the fact that different views of the plant may be needed in different future applications. It is not known today which will be successful, but we want to give the opportunity for both to be investigated. During the second year of the project, Testbed UAV will also be used.