Biomathematics & Statistics Scotland
Traditionally, phenotyping has required an enormous amount of human effort to measure plant characteristics. However, digital image analysis has the potential to automate the measurements that can be taken from whole plants or plant parts, saving time and increasing precision.
In collaboration with SASA, we have developed methods to be used in Distinctness, Uniformity and Stability Testing of new plant varieties. For example, dentation is an important feature of the shape of pea stipules, which can be removed from plants, photographed under laboratory conditions and then analysed. We have found that measurements, such as the frequency and shape of individual dentations in pea stipules, correlate well between automatic and manual scores.
 |
 |
Laboratory image of single pea (Pisum sativum) stipule. Dentation sharpness is defined as the angle between the two red lines, which are fitted automatically to images. |
Association between manual and automatic measures of sharpness of dentation for 23 pea cultivars. The manual score is based on a subjective visual scale, whereas the automatic method averages (10 * (1-cosine(angle))) over many dentations in many stipules. |
In an EU-FP7 project, with several partners including Wageningen University, The Netherlands, we are developing methods for phenotyping growing plants in greenhouse conditions. The project, called SPICY (Smart tools for Prediction and Improvement of Crop Yield), aims to develop an advanced suite of tools for breeding of crop plants, using pepper as the exemplar crop. The crucial first step in the image analysis, before any measurements can be taken, is to infer the distances of plants from the camera, using multiple views from different angles. This is a challenging task for complex images. The best results obtained so far use linear discriminant analysis to identify the background, then apply a standard stereo method to infer depths by minimising sums of squares of differences between small patches in the two images.
 |
Image of pepper plants |
 |
A pseudo-coloured depth map inferred from a stereo pair of images: objects closer to the camera are displayed in red, those further away are shown as shades of orange and yellow, and the background is labelled blue. |
Further details from: Graham Horgan and Yu Song
