Using hyperspectral imaging to characterize consistency of coffee brands and their respective roasting classes

Experimental data manipulations to assess performance of hyperspectral classification models of crop seeds and other objects

BACKGROUND

Optical sensing solutions are being developed and adopted to classify a wide range of biological objects, including crop seeds. Performance assessment of optical classification models remains both a priority and a challenge.

METHODS

As training data, we acquired hyperspectral imaging data from 3646 individual tomato seeds (germination yes/no) from two tomato varieties. We performed three experimental data manipulations: (1) Object assignment error: effect of individual object in the training data being assigned to the wrong class. (2) Spectral repeatability: effect of introducing known ranges (0-10%) of stochastic noise to individual reflectance values. (3) Size of training data set: effect of reducing numbers of observations in training data. Effects of each of these experimental data manipulations were characterized and quantified based on classifications with two functions [linear discriminant analysis (LDA) and support vector machine (SVM)].

RESULTS

For both classification functions, accuracy decreased linearly in response to introduction of object assignment error and to experimental reduction of spectral repeatability. We also demonstrated that experimental reduction of training data by 20% had negligible effect on classification accuracy. LDA and SVM classification algorithms were applied to independent validation seed samples. LDA-based classifications predicted seed germination with RMSE = 10.56 (variety 1) and 26.15 (variety 2), and SVM-based classifications predicted seed germination with RMSE = 10.44 (variety 1) and 12.58 (variety 2).

CONCLUSION

We believe this study represents the first, in which optical seed classification included both a thorough performance evaluation of two separate classification functions based on experimental data manipulations, and application of classification models to validation seed samples not included in training data. Proposed experimental data manipulations are discussed in broader contexts and general relevance, and they are suggested as methods for in-depth performance assessments of optical classification models.

Deep-Learning-Based Multispectral Image Reconstruction from Single Natural Color RGB Image—Enhancing UAV-Based Phenotyping

Multispectral images (MSIs) are valuable for precision agriculture due to the extra spectral information acquired compared to natural color RGB (ncRGB) images. In this paper, we thus aim to generate high spatial MSIs through a robust, deep-learning-based reconstruction method using ncRGB images. Using the data from the agronomic research trial for maize and breeding research trial for rice, we first reproduced ncRGB images from MSIs through a rendering model, Model-True to natural color image (Model-TN), which was built using a benchmark hyperspectral image dataset. Subsequently, an MSI reconstruction model, Model-Natural color to Multispectral image (Model-NM), was trained based on prepared ncRGB (ncRGB-Con) images and MSI pairs, ensuring the model can use widely available ncRGB images as input. The integrated loss function of mean relative absolute error (MRAEloss) and spectral information divergence (SIDloss) were most effective during the building of both models, while models using the MRAEloss function were more robust towards variability between growing seasons and species. The reliability of the reconstructed MSIs was demonstrated by high coefficients of determination compared to ground truth values, using the Normalized Difference Vegetation Index (NDVI) as an example. The advantages of using “reconstructed” NDVI over Triangular Greenness Index (TGI), as calculated directly from RGB images, were illustrated by their higher capabilities in differentiating three levels of irrigation treatments on maize plants. This study emphasizes that the performance of MSI reconstruction models could benefit from an optimized loss function and the intermediate step of ncRGB image preparation. The ability of the developed models to reconstruct high-quality MSIs from low-cost ncRGB images will, in particular, promote the application for plant phenotyping in precision agriculture.

Prediction of coffee aroma from single roasted coffee beans by hyperspectral imaging

•

This paper applied hyperspectral imaging (HSI) to predict roasted coffee aroma profile.

•

Individual roast coffee beans were analysed by HSI and aroma by GC–MS.

•

PLS models successfully predicted volatile aroma compounds in single coffee beans.

•

Beans were successfully segregated into two batches with different aroma profiles.

Coffee aroma is critical for consumer liking and enables price differentiation of coffee. This study applied hyperspectral imaging (1000–2500 nm) to predict volatile compounds in single roasted coffee beans, as measured by Solid Phase Micro Extraction-Gas Chromatography-Mass Spectrometry and Gas Chromatography-Olfactometry. Partial least square (PLS) regression models were built for individual volatile compounds and chemical classes. Selected key aroma compounds were predicted well enough to allow rapid screening (R² greater than 0.7, Ratio to Performance Deviation (RPD) greater than 1.5), and improved predictions were achieved for classes of compounds - e.g. aldehydes and pyrazines (R² ∼ 0.8, RPD ∼ 1.9). To demonstrate the approach, beans were successfully segregated by HSI into prototype batches with different levels of pyrazines (smoky) or aldehydes (sweet). This is industrially relevant as it will provide new rapid tools for quality evaluation, opportunities to understand and minimise heterogeneity during production and roasting and ultimately provide the tools to define and achieve new coffee flavour profiles.

Near-infrared spectroscopy and multivariate calibration as an alternative to the Agtron to predict roasting degrees in coffee beans and ground coffees

Agtron method is widely used in the industry to determine roasting degrees in whole and ground coffee but it suffers from some inconveniences associated with unavailability of equipment, high cost, and lack of reproductive results. This study investigates the feasibility to determine roasting degrees in coffee beans and ground specialty coffees using near-infrared (NIR) spectroscopy combined with multivariate calibration based on partial least squares (PLS) regression. Representative data sets were considered to cover all Agtron roasting profiles for whole and ground coffees. Proper development of models with outlier evaluation and complete validation using parameters of merit such as accuracy, adjust, residual prediction deviation, linearity, analytical sensitivity, and limits of detection and quantification are presented to prove their performance. The results indicated that predictive chemometric models, for intact coffee beans and ground coffee, could be used in the coffee industry as an alternative to Agtron, thus digitalizing the roasting quality control.

Compressive spectral imaging via deformable mirror and colored-mosaic detector

Compressive Spectral Imaging (CSI) is an emerging technology that aims at reconstructing a spectral image from a limited set of two-dimensional projections. To capture these projections, CSI architectures often combine light dispersive elements with coded apertures or programmable spatial light modulators. This work introduces a novel and compact CSI architecture based on a deformable mirror and a colored-filter detector to produce compressive spatio-spectral projections without the need of a grating or prism. Alongside, we propose a tensor-based reconstruction algorithm to recover the spatial-spectral information from the compressed measurements. Experimental results on both simulated and real datasets demonstrate efficacy of the proposed acquisition architecture and the especially crafted inversion algorithms.

Penetration and scattering-Two optical phenomena to consider when applying proximal remote sensing technologies to object classifications

Proximal remote sensing is being used across a very wide range of research fields and by scientists, who are often without deep theoretical knowledge optical physics; the author of this article falls squarely in that category! This article highlights two optical phenomena, which may greatly influence the quality and robustness of proximal remote sensing: penetration and scattering. Penetration implies that acquired reflectance signals are associated with both physical and chemical properties of target objects from both the surface and internal tissues/structures. Scattering implies that reflectance signals acquired from one point or object are influenced by scattered radiometric energy from neighboring points or objects. Based on a series of laboratory experiments, penetration and scattering were discussed in the context of "robustness" (repeatability) of hyperspectral reflectance data. High robustness implies that it is possible to control imaging conditions and therefore: 1) obtain very similar radiometric signals from inert objects (objects that do not change) over time, and 2) be able to consistently distinguish objects that are otherwise highly similar in appearance (size, shape, and color) and in terms of biochemical composition. It was demonstrated that robustness of hyperspectral reflectance data (40 spectral bands from 385 to 1024 nm) were significantly influenced by penetration and scattering of radiometric energy. In addition, it was demonstrated that the influence of penetration and scattering varied across the examined spectrum. Characterization of how optical phenomena may affect the robustness of reflectance data is important when using proximal remote sensing technologies as tools used to classify engineering and biological objects.

Hyperspectral imaging to characterize plant-plant communication in response to insect herbivory

BACKGROUND

In studies of plant stress signaling, a major challenge is the lack of non-invasive methods to detect physiological plant responses and to characterize plant-plant communication over time and space.

RESULTS

We acquired time series of phytocompound and hyperspectral imaging data from maize plants from the following treatments: (1) individual non-infested plants, (2) individual plants experimentally subjected to herbivory by green belly stink bug (no visible symptoms of insect herbivory), (3) one plant subjected to insect herbivory and one control plant in a separate pot but inside the same cage, and (4) one plant subjected to insect herbivory and one control plant together in the same pot. Individual phytocompounds (except indole-3acetic acid) or spectral bands were not reliable indicators of neither insect herbivory nor plant-plant communication. However, using a linear discrimination classification method based on combinations of either phytocompounds or spectral bands, we found clear evidence of maize plant responses.

CONCLUSIONS

We have provided initial evidence of how hyperspectral imaging may be considered a powerful non-invasive method to increase our current understanding of both direct plant responses to biotic stressors but also to the multiple ways plant communities are able to communicate. We are unaware of any published studies, in which comprehensive phytocompound data have been shown to correlate with leaf reflectance. In addition, we are unaware of published studies, in which plant-plant communication was studied based on leaf reflectance.

Development of Noninvasive Classification Methods for Different Roasting Degrees of Coffee Beans Using Hyperspectral Imaging

This study aimed to develop an approach for quickly and noninvasively differentiating the roasting degrees of coffee beans using hyperspectral imaging (HSI). The qualitative properties of seven roasting degrees of coffee beans (unroasted, light, moderately light, light medium, medium, moderately dark, and dark) were assayed, including moisture, crude fat, trigonelline, chlorogenic acid, and caffeine contents. These properties were influenced greatly by the respective roasting degree. Their hyperspectral images (874–1734 nm) were collected using a hyperspectral reflectance imaging system. The spectra of the regions of interest were manually extracted from the HSI images. Then, principal components analysis was employed to compress the spectral data and select the optimal wavelengths based on loading weight analysis. Meanwhile, the random frog (RF) methodology and the successive projections algorithm were also adopted to pick effective wavelengths from the spectral data. Finally, least squares support vector machine (LS-SVM) was utilized to establish discriminative models using spectral reflectance and corresponding labeled classes for each degree of roast sample. The results showed that the LS-SVM model, established by the RF selecting method, with eight wavelengths performed very well, achieving an overall classification accuracy of 90.30%. In conclusion, HSI was illustrated as a potential technique for noninvasively classifying the roasting degrees of coffee beans and might have an important application for the development of nondestructive, real-time, and portable sensors to monitor the roasting process of coffee beans.

Identification of coffee bean varieties using hyperspectral imaging: influence of preprocessing methods and pixel-wise spectra analysis

Hyperspectral imaging was used to identify and to visualize the coffee bean varieties. Spectral preprocessing of pixel-wise spectra was conducted by different methods, including moving average smoothing (MA), wavelet transform (WT) and empirical mode decomposition (EMD). Meanwhile, spatial preprocessing of the gray-scale image at each wavelength was conducted by median filter (MF). Support vector machine (SVM) models using full sample average spectra and pixel-wise spectra, and the selected optimal wavelengths by second derivative spectra all achieved classification accuracy over 80%. Primarily, the SVM models using pixel-wise spectra were used to predict the sample average spectra, and these models obtained over 80% of the classification accuracy. Secondly, the SVM models using sample average spectra were used to predict pixel-wise spectra, but achieved with lower than 50% of classification accuracy. The results indicated that WT and EMD were suitable for pixel-wise spectra preprocessing. The use of pixel-wise spectra could extend the calibration set, and resulted in the good prediction results for pixel-wise spectra and sample average spectra. The overall results indicated the effectiveness of using spectral preprocessing and the adoption of pixel-wise spectra. The results provided an alternative way of data processing for applications of hyperspectral imaging in food industry.

Non-destructive analysis of sucrose, caffeine and trigonelline on single green coffee beans by hyperspectral imaging

Hyperspectral imaging (HSI) is a novel technology for the food sector that enables rapid non-contact analysis of food materials. HSI was applied for the first time to whole green coffee beans, at a single seed level, for quantitative prediction of sucrose, caffeine and trigonelline content. In addition, the intra-bean distribution of coffee constituents was analysed in Arabica and Robusta coffees on a large sample set from 12 countries, using a total of 260 samples. Individual green coffee beans were scanned by reflectance HSI (980–2500 nm) and then the concentration of sucrose, caffeine and trigonelline analysed with a reference method (HPLC-MS). Quantitative prediction models were subsequently built using Partial Least Squares (PLS) regression. Large variations in sucrose, caffeine and trigonelline were found between different species and origin, but also within beans from the same batch. It was shown that estimation of sucrose content is possible for screening purposes (R² = 0.65; prediction error of ~ 0.7% w/w coffee, with observed range of ~ 6.5%), while the performance of the PLS model was better for caffeine and trigonelline prediction (R² = 0.85 and R² = 0.82, respectively; prediction errors of 0.2 and 0.1%, on a range of 2.3 and 1.1% w/w coffee, respectively). The prediction error is acceptable mainly for laboratory applications, with the potential application to breeding programmes and for screening purposes for the food industry. The spatial distribution of coffee constituents was also successfully visualised for single beans and this enabled mapping of the analytes across the bean structure at single pixel level.

•

NIR hyperspectral imaging was applied to green coffee beans on a single seed basis.

•

Sucrose, caffeine and trigonelline were analysed from single beans using HPLC-MS.

•

PLS regression models were built to quantify for the coffee constituents.

•

Prediction models were applied to single pixels to visualise the compound distribution.

•

HSI was shown to be useful for rapid non-destructive screening of single coffee beans.

Using proximal remote sensing in non-invasive phenotyping of invertebrates

Proximal imaging remote sensing technologies are used to phenotype and to characterize organisms based on specific external body reflectance features. These imaging technologies are gaining interest and becoming more widely used and applied in ecological, systematic, evolutionary, and physiological studies of plants and also of animals. However, important factors may impact the quality and consistency of body reflectance features and therefore the ability to use these technologies as part of non-invasive phenotyping and characterization of organisms. We acquired hyperspectral body reflectance profiles from three insect species, and we examined how preparation procedures and preservation time affected the ability to detect reflectance responses to gender, origin, and age. Different portions of the radiometric spectrum varied markedly in their sensitivity to preparation procedures and preservation time. Based on studies of three insect species, we successfully identified specific radiometric regions, in which phenotypic traits become significantly more pronounced based on either: 1) gentle cleaning of museum specimens with distilled water, or 2) killing and preserving insect specimens in 70% ethanol. Standardization of killing and preservation procedures will greatly increase the ability to use proximal imaging remote sensing technologies as part of phenotyping and also when used in ecological and evolutionary studies of invertebrates.