For plant variety testing many plant characteristics need to be measured to assess distinctness between varieties and uniformity and stability of a variety. The characteristics are described in guidelines and protocols for each crop, e.g. wheat, onion or carrot. These guidelines are established by the International Union for the Protection of New Varieties of Plants (UPOV, www.upov.int) and the Community Plant Variety Office (CPVO, www.cpvo.europa.eu). Several characteristics, related to shape and form are tedious to measure by hand and should be measured in an objective and reproducible way. The use of image analysis leads to significant decrease of time that is needed to make observations and scores the characteristics. Moreover, the results obtained by using image analysis are much more accurate than results of manual measurements.
In the Netherlands, a user-friendly image analysis system has been established for recording and analysis of images. For image recording a standard slr camera was used. The camera is mounted on a repro stand with front and back illumination. The NKRemote software package (Breeze systems) controls the camera and images are captured and saved for offline processing.
The recorded images are then analyzed with ImageJ. For this an ImageJ plugin was developed with a user-friendly interface, based on ‘Action Bar’ by Jerome Mutterer. The amount of parameters and settings for the user are kept to a minimum. Currently the following plant parts/characteristics are implemented:
- Flax seeds: length and width based on the fitting ellipse.
- Flax-bolls: length and width based on the minimum-enclosing rectangle in the direction of the peduncle.
- Pods of pea and beans: length, width and bending, based on the skeleton and the size of the beak. For the skeleton the shortest path algorithm of the ‘analyze skeleton’ plugin is used.
- Sugar beet cotyledon: length, width and area based on the distance transform.
- Carrot root: length, width, width of the crown and shape factor based on the inflection points of the outer contour.
Calibration is done using a calibration disc in a separate part of the image on a blue background for automatic processing. Identification of the images is done using a QR-Code label in the same blue part.
The software is proven to be very robust, where disturbing factors on the objects, such as small crown leaves are removed automatically.
We conclude that ImageJ as popular open source imaging platform is a good choice for implementing an image analysis system for plant variety testing to automatically and reliably measure a large variety of plant characteristics.
shape-features, calibration, labeling, user-friendly,
Gerrit Polder (1963) is researcher Image Analysis at the department Biometris of Wageningen University and Research Center. In 1985 he obtained a B.Sc. in electronics at the HAN University of Applied Sciences in Arnhem the Netherlands. After working several years in image processing and related topics, in 2004 he got a Ph.D. from Delft University of Technology on Spectral imaging for measuring biochemicals in plant material. From 2004 he works at the Biometris group on machine-vision and robotics projects for disease detection in agricultural fields using hyperspectral and multispectral camera systems, automated phenotyping in greenhouses using stereo-vision and Time Of Flight Imaging, sensor fusion (color, fluorescence and infrared) for monitoring plant health using a robot system, and other projects mainly focused on agricultural research.
Presenting author: Gerrit Polder
Organisation: Wageningen UR (University & Research centre)
co-authors: Gosia Blokker
Gerie W.A.M. van der Heijden.