Extracting the data from digital phenotyping images
As I explained in my earlier articles, modern phenotyping is often image based and involves a range of different kinds of sensor technologies. Often, the sensors are cameras, but crucially, they are cameras with different wavelength sensitivities.
First you produce the images and then you analyse those images to extract particular features. For example, when you take an image of a plant it will have some leaves on it, perhaps it will have some flowers, and there might be other things, such as the soil or whatever growing media is being used. You can separate out those different parts of the plant within the image and then – if you’ve got different wavelength information associated with the image – you can extract that wavelength image from the different plant parts.
This, actually, is the basis of the kinds of facilities that we have both at Cranfield and Rothamsted. They might take, first of all, regular RGB images: and you can extract a lot of information from an RGB image because you can measure things like height and shape and form and growth stage or disease incidence, for example.
However, if you need to look for things like disease incidence, then maybe it’s not something that a regular RGB camera, or the eye, can see. Here the application of thermal camera, multi and hyper spectral cameras and fluorescence detectors become important.
One consequence is that these approaches result in massive data sets that need appropriate handling algorithms and safe storage. It is also important that access to the datasets is facilitated.
This is where proxy measurements come in. Information derived from the different spectral sensors enable estimation of parameters which are not immediately obvious. Perhaps something which isn’t quite what you want, but which is telling you about what you do want.
Some specific wavelengths might have greater sensitivity to the particular features you are looking to measure, maybe it’s florescence, maybe it’s thermal imagery or any other specific wavelength, rather than the whole visible spectrum. If you focus on a particular wavelength then you might be able to pick out specific features. And both the Field Scanalyzer here at Rothamsted and the phenotyping gantry in the Soil Health Facility at Cranfield have these sorts of capabilities, as does the equipment in CHAP’s Digital Phenotyping Lab.
Next time I will look at how the data collected by all these facilities can be used to help scientists gain greater insights into plant health.
Read the previous articles in this series:
In the concluding part of this series, Developing proxies through digital phenotyping, Professor Hawkesford will look at how scientists are constantly working to align the spectral signatures of images with parameters to indicate changes.