Colin Victor. Greensill

Sorting of Fruit Using near Infrared Spectroscopy: Application to a Range of Fruit and Vegetables for Soluble Solids and Dry Matter Content:

The performance of a single instrumentation platform, incorporating the use of a tungsten halogen light source, body transmittance optics and a silicon photodiode array detector, and a uniform chemometric approach is reported for the application of assessment of determination of soluble solids and dry matter content of a range of fruit. Spectra were acquired at integration times of 30 ms or less, with integration time varied between fruit types to achieve a similar signal level. Calibration performance was compared in terms of root mean standard error of cross validation (RMSECV), regression coefficient (R), and the SDR (SDR = SD / RMSECV (SD is standard deviation)]. The technology was well suited to sorting on soluble solids content (SSC) in apple (RMSECV 0.22%, SDR > 5; R 0.98), and useful, in decreasing order of accuracy, for sorting of stonefruit, mandarin, banana, melons, onions, tomato and papaya (RMSECV 1.1%, SDR 1.6, R 0.79). The technology also performed well in sorting on dry matter content in kiwifruit (RMSECV 0.38%, SDR > 3, R 0.95), and useful, in decreasing order of accuracy, for sorting of banana, mango, avocado, tomato and potato (RMSECV 1.0%, SDR 1.7, R 0.79). The limitations of the application of the technology to fruit sorting is discussed in terms of fruit type (“skin” thickness) and population range. For example, calibration RMSECV was only 0.20% on tomato SSC, but as population variation was low (SD 0.30%), a poor R (0.77) and SDR (1.5) was obtained.

Calibration transfer between miniature photodiode array-based spectrometers in the near infrared assessment of mandarin soluble solids content

The transfer of predictive models among photodiode array based, short wave near infrared spectrometers using the same illumination/detection optical geometry has been attempted using various chemometric techniques, including slope and bias correction (SBC), direct standardisation (DS), piecewise direct standardisation (PDS), double window PDS (DWPDS), orthogonal signal correction (OSC), finite impulse transform (FIR) and wavelet transform (WT). Additionally, an interpolation and photometric response correction method, a wavelength selection method and a model updating method were assessed. Calibration transfer was attempted across two populations of mandarin fruit. Model performance was compared in terms of root mean squared error of prediction (RMSEP), using Fearns significance testing, for calibration transfer (standardisation) between pairs of spectrometers from a group of four spectrometers. For example, when a calibration model (Root Mean Square Error of Cross-Validation [RMSECV = 0.26% soluble solid content (SSC)], developed on one spectrometer, was used with spectral data collected on another spectrometer, a poor prediction resulted (RMSEP = 2.5% SSC). A modified WT method performed significantly better (e.g. RMSEP = 0.25% SSC) than all other standardisation methods (10 of 12 cases), and almost on a par with model updating (MU) (nine cases with no significant difference, one case and two cases significantly better for WT and MU, respectively).

Sugar "imaging" of fruit using a low cost charge-coupled device camera

Sugar “imaging” of fruit has previously been reported using NIR filters and relatively expensive (high signal-to-noise) charge-coupled device (CCD) instrumentation. In a bid to use lower cost CCD instrumentation (criterion of less than AU

Assessment of quality defects in macadamia kernels using NIR spectroscopy

5,000 total component costs), the signal-to-noise constraint on calibration model performance was investigated by artificially degrading spectra from a 15-bit AtoD system. A low cost 8-bit CCD camera was then used in conjunction with a filter wheel in a transmittance configuration employing three 50 W halogen lamps. Multiple linear regregression calibrations were developed based on absorbance data of five wavelengths (830, 850, 870, 905 and 930 nm) relevant to sugar and water. Calibration models for the sucrose concentration of solutions on a cellulose matrix were poor (R2 = 0.4) when based on individual pixel data, but acceptable (R2 = 0.98, RMSECV = 1.1) (n = 20, mean = 13.9% total soluble sugars (TSS), SD = 6.04) when based on an average of a 23 × 23 pixel block (i.e. 529 pixels). For a calibration based on melon tissue TSS, using spectral data averaged over groups of 529 pixels, results were poorer than expected (R2 = 0.4, RMSEP = 1.74 (n = 163, mean = 9.45, SD = 2.07% TSS). Predicted TSS output for all pixel blocks from an image was used to generate a false colour image. We conclude that this application requires a higher level of signal-to-noise (for example, 10-bit, > 60 dB CCD).

Temporal and environmental sensitivity of a photodiode array spectrophometric system

Spectral data were collected of intact and ground kernels using 3 instruments (using Si-PbS, Si, and InGaAs detectors), operating over different areas of the spectrum (between 400 and 2500 nm) and employing transmittance, interactance, and reflectance sample presentation strategies. Kernels were assessed on the basis of oil and water content, and with respect to the defect categories of insect damage, rancidity, discoloration, mould growth, germination, and decomposition. Predictive model performance statistics for oil content models were acceptable on all instruments (R2 > 0.98; RMSECV 0.97; RMSECV < 0.2%), whereas calibration models for intact kernels were relatively poor. Calibration coefficients were more highly weighted around 1360, 740 and 840 nm, consistent with absorbance due to overtones of O-H stretching and combination. Intact kernels with brown centres or rancidity could be discriminated from each other and from sound kernels using principal component analysis. Part kernels affected by insect damage, discoloration, mould growth, germination, and decomposition could be discriminated from sound kernels. However, discrimination among these defect categories was not distinct and could not be validated on an independent set. It is concluded that there is good potential for a low cost Si photodiode array instrument to be employed to identify some quality defects of intact macadamia kernels and to quantify oil and moisture content of kernels in the process laboratory and for oil content in-line. Further work is required to examine the robustness of predictive models across different populations, including growing districts, cultivars and times of harvest.

Comparison of two NIR systems for quantifying kaolinite in weipa bauxites

The effect of the spectral variation in quartz tungsten lamp output with respect to elapsed time from power-up and variation in environmental temperature, and the variation in readout in the front-end electronics (FEE) and spectrometer with temperature, on predictive model performance of total soluble solids (TSS) in intact fruit was assessed for a silicon photodiode spectrometer-based system. Lamp (10 each of OSRAM HLX64623 and Sylvania 521995 12 V 100 W GY6.35 quartz tungsten halogen) output was assessed at 10 s intervals over a 4 h period, and 10 min intervals over approximately 3000 h. The environmental temperature of each component in a near infrared (NIR) spectroscopy system (lamp, FEE, spectrometer) was incrementally adjusted in 10°C intervals between 10°C and 60°C. The lamp output was spectrally stable within the time of the first measurement (10 s), although total illumination was not stable until approximately 40 min from start-up. Thus, the performance of the predictive models based on second derivative of absorbance data was not significantly impacted by lamp warm-up time. Noise on measurement associated with the use of a single white reference resulted in a mean increase in root mean square error of prediction (RMSEP) as high as 0.22% TSS and individual increases as high as 0.82%. Averages of white reference measurements significantly improved performance. When predictive models were developed using second derivative absorbance data and averaged (10) white references, there was no statistically significant impact in RMSEPs on time of lamp warm-up (after 10 s), even during the last hours of lamp life. Spectral variation resulting from changes of NIR system components (lamp and FEE) also affected lamp output quantity rather than quality and thus did not affect the predictive performance owing to the second derivative absorbance pretreatment. Some lamps displayed start-up output characteristics on their first use that were not repeated in subsequent trials. This result indicates the need for a short lamp “burn-in” period.

Impact of resolution on NIR PLS calibration of kaolinite content with Weipa bauxite

Two laboratory-based near infrared systems were compared for the purpose of quantifying kaolinite within Weipa bauxites. The two systems were based on different instrument technologies, a Bruker MPA (multi-purpose analyser) FTNIR (Fourier transform near infrared) instrument, and a Foss Model 6500 pre-dispersive grating instrument. Systems were compared in terms of accuracy, robustness, sample throughput and ease of use. The MPA-based system was superior in terms of accuracy (root mean squared error of prediction (RMSEP) = 0.88%m/m), throughput and ease of use. The Model 6500 had an advantage of offering the most robust measurement with greater time stability and lower errors in repeated and repacked measurements. However, addition of a small number of temporally displaced and repacked samples in the calibration set improved robustness of the MPA system, matching the Model 6500.

Light-emitting diodes as light sources for spectroscopy: Sensitivity to temperature:

The impact of spectral resolution on the quantification of kaolinite in Weipa bauxite was assessed with a view to the development of a quantitative tool for routine onsite use. Spectral data (calibration set = 204 samples, prediction set = 60 samples) were acquired at five resolution settings (2, 4, 8, 16 and 32 cm−1) using a Bruker MPA Fourier transform NIR (FTNIR) instrument. Visually, the increase in resolution from 32 cm−1 to 8 cm−1 yielded an apparent increase in spectral feature definition whilst further increase in resolution (4 or 2 cm−1) gave no additional benefit. PLS models, using identical data-treatments and spectral windows, were developed for each resolution setting. Root mean squared errors of prediction (RMSEP) were not significantly different for the resolutions 2, 4, 8 and 16 cm−1 whilst models built at the 32 cm−1 had RMSEPs significantly 4–5% worse than the other resolutions. The results demonstrated that PLS calibration of kaolinite content in Weipa bauxite was nominally affected by resolution with no predictive gain demonstrated beyond 16 cm−1.

Improving calibration transfer between shortwave near infrared silicon photodiode array instruments

Understanding of light-emitting diode lamp behaviour is essential to support the use of these devices as illumination sources in near infrared spectroscopy. Spectral variation in light-emitting diode peak output (680, 700, 720, 735, 760, 780, 850, 880 and 940 nm) was assessed over time from power up and with variation in environmental temperature. Initial light-emitting diode power up to full intensity occurred within a measurement cycle (12 ms), then intensity decreased exponentially over approximately 6 min, a result ascribed to an increase in junction temperature as current is passed through the light-emitting diode. Some light-emitting diodes displayed start-up output characteristics on their first use, indicating the need for a short light-emitting diode ‘burn in’ period, which was less than 24 h in all cases. Increasing the ambient temperature produced a logarithmic decrease in overall intensity of the light-emitting diodes and a linear shift to longer wavelength of the peak emission. This behaviour is consistent with the observed decrease in the IAD Index (absorbance difference between 670 nm and 720 nm, A670–A720) with increased ambient temperature, as measured by an instrument utilising light-emitting diode illumination (DA Meter). Instruments using light-emitting diodes should be designed to avoid or accommodate the effect of temperature. If accommodating temperature, as light-emitting diode manufacturer specifications are broad, characterisation is recommended.

A remote acceptance probe and illumination configuration for spectral assessment of internal attributes of intact fruit

The use of a model developed on spectra of one (master) instrument with spectra collected using another (slave) instrument requires differences in spectra of master and slave units to be orthogonal to the calibration model. The more spectral similarity is achieved in hardware, i.e. by matching the optical characteristics of the devices, the less chemometric correction is required. The transfer of partial least squares models for total soluble solids (TSS) of intact apple fruit between instrumentation based on silicon photodiode arrays was improved by use of more accurate wavelength assignments over the wavelength range used in the model. Several transfer methodologies were trialled, including piecewise direct standardisation (PDS), transfer by orthogonal projection, model updating (MU) and difference spectrum adjustment. The difference spectrum method combined with new wavelength assignments and MU gave results comparable to the performance of the master instrument and to models directly developed on the slave instruments (r2 = 0.95, SEP-b = 0.47 and bias = −0.03% TSS, for a population of mean = 14.45 and SD = 1.64% w/v). The use of average difference spectrum adjustment combined with MU was preferred over PDS because of ease of implementation.