A spectral imaging system for detection of Botrytis in greenhouses

The healthy greenhouse

In the Interreg IV, EU project 'The healthy greenhouse' a new integral crop protection system is developed. The total system consists of equipment and software for sustainable management for modern horticultural companies. One of the products is an application-specific multispectral camera for detection of fungal diseases, specifically Botrytis on Cyclamen.

Application-specific multispectral camera.

The ultimate goal is to develop a fast multispectral camera using PixelTec™ technology in three steps:

1. Diseased and healthy plants are imaged in the lab using a slow hyperspectral system with 185 spectral bands. The most discriminating bands are selected using pattern recognition techniques.   
   
2. Validation in the greenhouse with a fast filter wheel based system using the bands found in the first step.
3. On sensor implementation using PixelTec™ technology for an application-specific camera.

Experiment

In the lab infected and healthy plants were imaged using the hyperspectral camera. Supervised classification using linear discriminant analysis was applied to detected Botrytis. The results show very good localization of infected spots. These labeled images are used as ground truth for selecting the most discriminating wavelengths for filter selection for the multispectral camera.

Results

A feature selection algorithm was applied to find the eight most discriminating bands. These bands were validated on pixels, which were not used in the feature selection step. 90% of the pixels were properly classified, which indicated that the selected bands are suitable for Botrytis selection in Cyclamen. The selected bands will be validated in a greenhouse experiment using a fast filter wheel camera and finally implemented using PixelTec™ technology.