Hyperspectral examples

Agestructure of vine leaves in hyperspectal terms — Hyperspectral signatures with varying leaf ages

Examples of hyperspectral signatures of different types of tissue and related model fitting

In most cases spectral measurements are done on green leaves or canopies. The green tissue generates a characteristic signature pattern, which varies with the age structure of the plant. In fact, the two dynamic processes "growth" and "senescence" can mask any other factors. A fully grown, mature leaf does not change much, we see some kind of upper asymptote in the signature, again, masking potential traits or factors of interest. But the library also shows the diversity of hyperspectral signatures and gives reasons affecting the signatures. The visual comparison with model fits demonstrate the flexibility of our analysis method for all available spectra, transfering hyperspectral signals into parameters usable for statistics and comparisons.

Classical standard spectrum of a mature green leaf with all characteristics of a sugar beet leaf. (leaf structure, cuticula, leaf diameter etc.). Example demonstrates the asymptotic boundary of a healthy, mature crop. Measurement taken above canopy.

Again, one of the classical results: green, mature sugar beet leaf, measurement taken with plant probe directly taken from the leaf surface with own light source and no disturbances in the signal. Measurements taken with a plant probe were found to result in stable signals and are most likely to represent some kind of "true" signal of a tissue. we have constant light conditions and no disturbances in the reflectance. Data gained on this base allow the statistical analysis with respect to minor differences in the parameters. We recommend this method for specific questions in stress responses or trait phenotyping in greenhouse experiments or simple pot experiments, etc.

Example of an N fertilised vine leaf of elder age, but before the beginning of senescence.

Vine leaf in the beginning of senescence stage and disturbances in some domains (760nm). Fitting the model adjusts such disturbances without loss of information. That is of need, as the usual analysis methods might emphasise on that domain and would find arbitrary differences.

Green leaf with a thick cuticula and waxed leaf surface, strong light absorption (= low reflection) in the chlorophyll domains.

Sunflower leaf, no differences compared to vine leaves on the spectral scale, soft, hairy leaf.

Demonstration model of a phenotyping facility with immediate spectral analysis (see background screen). Example of crop measurement, when both plant morphology and treatment (fertiliser) affect the result and disturb the hyperspectral reflectance.

Example: sensing a barley plant in a pot experiment, reflectance measurement from above. Resulting deviance and and disturbances are due to the morphology of the plant and imperfect artificial light conditions, leading to oscillation in the sensor signal. Smoothed by model fit, treatment: compound fertiliser.

Example as above, treatment no potassium (no K)

Example as above: no nitrogen, poorly developed leaves, the reflectance of the soil predominates the signal.

Example as above: no phosphor

Example as above: no sulphur.

as above, no zinc

Chestnut leaf, dead, brown tissue, any structure in the signal is lost.

Chestnut leaf at the start of senescence, chlorophyll partly transferred, necrotic spots on leaf surface.

Measurement taken from another organ, grapes at different stages of maturity with respect to nitrogen supply.

Reflectance of grapes with Botrytis infestation.

Spectrum of grapes with nitrogen supply.

Grapes with advanced botrytis mycel due to higher nitrogen level. The pathogen has already destroyed the tissue structure of the grape.

Example of an imaging line scanner. The averaged spectrum is equal to a non-imaging sensor, but the spectrum of one pixel shows numerous scatters and disturbances over all domains. Here is the model fitting of particular importance to smooth the signature, as analysis methodes might stress the scatter in single domains instead the true treatment effect.

Same object and problem as above, but signature taken from a pixel representing healthy leaf tissue.

Measurement taken from a green fruit instead of leaves. Another plant organ with different tissue structure, waxed surface and high water content. Analysis of such patterns again by model fitting.

Measurement from a red tomato fruit. The chlorophyll related domains are changed by the red pigments. Fitting and analysis by model application.

Example for spectra up to 2500nm, measured with contact probe, means the common perturbations of the water domains are missing. Model fitting and analysis include information from the SWIR domains. As the model has much more parameters compared to a fit up to 1050nm, much larger sample sizes are required.

Canopy measurement with light adaption up to 1600nm. With respect to exposure time we see perturbations and jumps in the signature plus the problems with the water domains. Model fit smooth both jumps and perturbations.

As above, but with contact probe, omitting many sources of error. Model fit up to 1600 nm.

Spectrum of a NIRS spectrometer in the range from 1200 to 2400 nm; used for the estimation of ingredients and substances in prepared samples of any tissue. Can be used to accelerate analytical processes in the lab or to replace parts of complex processes by NIRS information. Establishing sufficient sample size can be used for the calibration of empirical correlations of spectra to the amount of substances. It does work for most of components, not only for dry matter or protein contents or similar.

Reflectance of the skin. Spectra vary with numerous factors as melanin content, skin humidity, fat, age, etc. All those spectra can be fitted to the Weibull model, classification are possible with large samples.

Mean spectra of the skin including a liver spot. The size of scanned area by the contact probe is too large to give evidence of some anomaly.

Human tissue (oral mucosa) with high humidity level.

Minced pork meat, decayed, (red square) with obvious colour changes and high CFU values.

Frisches Hackfleisch und Plant Probe Messung

Fresh minced meat and its characteristic spectrum, red colour might be increased due to atmospheric packages. Hyperspectral reflectance measurements can be used for quality management and production control in slaughterhouses.

Minced meat stored after one week at 4°C. The meat has changed visually and also its consistency and the reflectance signature.

Pork cutlet signature with contact probe; the sample is visually red, no oxidation has occurred yet. The CFU values are in the range of (log) 4. For comparison the spectrum of a decayed cutlet sample is shown in the graph. The signature represents CFU values >8 to 9.