Hummingbird is a market leading imagery analytics provider.
Predictive Analytics is traditionally depicted as a timeline. What has happened in the past, what is happening right now in the present and with predictive analytics, what is going to happen in the future. Our Data Scientists use a range of tools and techniques to answer these questions.
Our Data Scientists talk about our work in Predictive Analytics
Field Boundary Detection
Automatic detection of field boundaries from satellite images. It uses Sentinel-2 images (10 m resolution) in 4 ranges of the electro-magnetic spectrum (red, green, blue and near-infrared) as an input. A deep learning model (ResUnet-a) has been trained on thousands of annotated images to learn how to identify field boundaries, reaching an accuracy of ~90%
Crop Type Classification
Hummingbird Technologies has developed a Crop Type Classification tool that uses state-of-the-art machine learning techniques to provide monthly in-season crop type predictions. The model uses the temporal evolution of the observed satellite data as the season progresses to identify crops. The model currently uses Sentinel-1 backscatter data as a base input to guarantee predictions can be made throughout the season regardless of cloud cover, and fuses this input data with Sentinel-2 reflectance data
This product provides an in-season prediction for expected yield at varying spatial resolutions, ranging from regional to national level. Our predictions are derived by using parameters such as historic yield information, soil and weather data to calibrate advanced phenological crop growth models. Data assimilation techniques are then used to further improve accuracy by constraining the numeric output of the crop model with Leaf Area Index (LAI) values observed from satellite.