With the help of embedded systems we create labelled data for applications in which machine learning meets agriculture. Good ML-models needs good data. ALITA and EAGL-I have already generated hundreds of thousands of labelled plant images, ranging from vegetables, over weeds, to typical crops found in the Prairies such as Canola and Soybeans. We keep imaging and soon extend our data acquisition to hyperspectral and 3d-imaging.
We use high performance GPU-servers to train deep neural networks on our image data. Tasks range from simple plant identification, classifying a plant as crop or weed, or even evaluating a plant's well-being. These machine learning models train on the data we produced ourselves to be the "brains" in future robotic agents taking care of fields one plant at a time.
All our data is stored and is available for researchers and eventually to everyone online. We have developed a front-end client that allows everyone to browse and filter the provided data to get exactly what they need for their application's development. The client can be found here. This publicly available data will accelerate the research community and the agriculture industry towards the next revolution in how we grow our food.