Machine learning for automatic stress detection in horticultural production

Important results from the project

Our method does not seem to be sensitive enough to detect biotic stress (powdery mildew on strawberries) at an early stage. The method is an ML-based analysis of step response parameters of chlorophyll fluorescence, which has proven promising for other types of stress factors. The method includes an ML algorithm for leaf segmentation that has been developed and is also interesting for other purposes. We have also evaluated and proceeded with an ML-based tool for sunlight forecasting.

Expected long term effects

Hopefully, the fluorescence method can find other applications, such as detecting other types of biotic stress as well as leaf segmentation. The improved sunlight forecasting is something that will be utilized by our customers to optimize lamp use, save energy, and reduce electricity cost. The project also contributes to follow-up projects and important collaborations within the ´smart farming´ segment.

Approach and implementation

Resources from Heliospectra and Chalmers have focused on the analysis of time- and space-resolved data from a fluorescence camera. The dataset was generated outside the project. Available resources changed somewhat during the course of the project, and as a result, the project operated at a slower pace for a longer period to make the best use of the funds. When the main result was negative, we were able to redirect the project to test ML tools for another application, which was rewarding.