Silver mirror

Reflectance spectra of the fruit surface were measured on a microscopic scale (spot size: ≈10 µm) which allowed the collection of optical signals from individual cells. The halogen lamp of the microscope served as light source in bright-field configuration. Light reflected from the sample passed back into the objective and was coupled in confocal configuration with a 100 m core optical fibre connected to a spectrometer (QE65000, Ocean Optics, 200–880 nm). The reflection spectra were normalized with respect to a silver mirror (Thorlabs). Spectra and images were collected using unpolarized illumination and a circularly polarizing filter consisting of a superachromatic quarter waveplate (B. Halle) combined with a liner polarizer (Thorlabs) for right-handed (RH) and left-handed (LH) light detection. The hyperspectral images were collected in the same configuration using an additional liquid crystal filter (CRI, Varispec) that was inserted in front of the CCD imaging chip. Images were collected with a camera and carefully normalized if recorded with different exposure times, considering also the nonlinearity of the camera response.

A PARISS^® hyperspectral imaging system (LightForm, Inc., Asheville, NC) was used to provide spatial mapping of the spectral output from the samples when normally illuminated with white light. Each sample was imaged without a coverslip for structure and spectral mapping. For structure 20% of the light output was imaged using a monochrome QIClick camera (QImaging); for spectral reflectance measurements, 80% of the light output was collected using a 100× 0.9 N.A. air objective (giving ≤ 0.5 µm spatial resolution) on a Nikon Eclipse 80i microscope with a PARISS^® spectrometer utilizing a Retiga 2000DC CCD camera (QImaging). A 50 µm slit was used for window collection, and radiometric calibration was done with a Hg⁺/Ar⁺ lamp (LightForm Inc.), with spectral resolution measured better than 2 nm. A silver mirror (Thorlabs) was used as a reflectance reference for all measurements. Hyperspectral mapping was performed using a library of selected spectra with a minimum correlation coefficient (MCC) of 0.99 used as a discrimination factor to identify and map common spectra. All spectra were smoothed, normalized, and plotted using GraphPad Prism statistical software (GraphPad Software, Inc., La Jolla, CA, USA). The colours of the curves were estimated based on the “spec2rgb” function in R script “pavo”⁴⁹ .