Hyperspectral Imaging of Plants
Hyperspectral imaging is a technology widely used for remote imaging, in an effort to extract a maximum of information out of images made under strongly varying imaging conditions, caused by the variability of the sunlight and changing atmospheric conditions. On the other end of the spatial scale, Fourier Transform spectroscopy of biological material is used to assess specific ingredients contained in samples that are often grinded and dried, and thus show no spatial resolution at all. The application of hyperspectral imaging under highly controlled illumination conditions, as is standard in all LemnaTec scanalyzer3D imaging stations, provides new options for data generation with high spectral resolution in a certain range of the full electromagnetic spectrum of frequencies. In contrast to the LemnaTec usage of multiple frequency imaging (VIS, NIR IR), hyperspectral imaging is rather focused on a smaller range (e.g. 400–1000 nm), but takes images at a spectral resolution between 1 and 10 nm. The user is thus enabled either to acquire full-spectrum datasets for each individual pixel of the image, or to restrict data acquisition to some specifically interesting frequency ranges. These will strongly depend on the substances, the substance groups or the general approach to hyperspectral data acquisition.
Hyperspectral imaging for in-depth analysis of spectral information is complemented by the high-throughput options that LemnaTec scanalyzer systems can provide.
Approaches to hyperspectral imaging using LemnaTec hyperspec technologies
If the absorption and reflectance spectra of substances like chlorophyll, anthocyans or others are identified, hyperspectral images can achieve at least a semi-quantitative value for substance concentration. To obtain reliable data, a larger validation experiment, including measurement of the values with other methods, is necessary in order to develop a concentration model that relates spectral information to concentration.
Learned finger printing or pattern approach
When an individual substance would only be used as a surrogate value for a more complex physiological phenotype, it may often make sense not to replicate the chemical measurement by hyperspectral imaging (substance-specific approach), but to try accessing the physiological phenomenon more directly. Based on a larger set of full-spectral information and a combination of specific measurement methods or plant pre-treatments (or plants with known specific backgrounds), a set of spectra is generated with a correlation to specific plant conditions such as biotic or abiotic stressors, senescence, nutrient deficiencies or different stages of ripeness. By employing direct comparison of spectra, advanced statistical analysis or machine-learning processes, different patterns and the spectral regions of interest for the discrimination are identified. These areas will then be monitored under routine conditions to assess details of the plant status.
Pure pattern finding approach
To make the best of screening, which means to look out for the unknown, it is useful to assess a certain spectral resolution for the entire spectrum and the whole plant. After the experiment, specific algorithms search for patterns or deviations from control plants or similarities to known, interesting plant types in the experiment. This open approach minimises the need for extensive calibrations and retains the flexibility to detect the truly new and innovative traits.