Automatic validation of sensor data
The project identifies common sensor problems for the most interesting sensors in agriculture, including sensors for temperature, humidity, precipitation, and wind in weather stations, sensors for soil temperature and humidity, etc.
We have developed models and algorithms (machine-learning, ML) and trained these with previously collected data from agriculture (eg via LantMät administered by Jordbruksverket).
Agriculture is about to become data-driven, and it is then necessary to qualify raw sensor values before they can be used as a decision basis in various models for plant protection, nutrient supply, harvest calculations, etc. There will be no possibility for people to overlook the massive amount of data that is generated monthly or annually.
Develop methods and models for Machine Learning
Modern ML-system (within ‘deep learning’) can now qualify content better than humans, e.g., for the review of images, ML systems create 2.5% errors on average, compared to 5% created by people. Such methods are used and manage multiple neighboring sensor systems to find errors and correct and replace incorrect data from individual sensors with estimated reasonable values, in a similar way that people act to validate data for calculation models to be run.