Hyperspectral imaging on food: data acquisition and processing tools ...

Interdepartmental Workshop on Hyperspectral Imaging University of Natural Resources and Life Sciences, Vienna Konrad‐Lorenz‐Straße 24, 3430 Tulln 20 March 2015 Within the frame of the FP7 SLOPE project (Integrated proceSsing and controL systems fOr sustainable forest Production in mountain arEas) and the BiRT (Bio‐ Resources & Technologies Tulln) Workshop Series.

Introduction and Welcome

We are very happy to welcome you at the 1st Interdepartmental Workshop on Hyperspectral Imaging at the University of Natural Resources and Life Sciences – BOKU-UFT linked to the SLOPE project (FP7- Collaborative Project – 604129; “Integrated proceSsing and controL systems fOr sustainable forest Production in mountain arEas”) and BiRT initiative (Bio- Resources & Technologies Tulln). Due to the excellent contributions by renowned scientists, who agreed to participate the workshop at minimum compensation for travel, stay and work, we were able to put together an exciting program, including an introductory chemometry course on hypercube processing. We hope that this workshop meets your expectations, opens up new perspectives and stimulates ideas for collaboration and development of joint research projects in the expanding and exciting field of Hyperspectral Imaging. We wish you an interesting workshop and a pleasant stay in Tulln! The BOKU organizers, Andreas Zitek Thomas Prohaska Barbara Hinterstoisser Katharina Böhm

SUPPORTED BY

Workshop on Hyperspectral Imaging | BOKU‐UFT

1

Programme 08:30 - 09:00 Registration and get together (Poster already on site) 09:00 - 09:15 Opening SLOPE and BiRT (Andreas Zitek, Barbara Hinterstoisser vice rector of BOKU, Rupert Wimmer - representative of the BiRT initiative) 09:15 - 10:00 Hyperspectral imaging – general introduction (Prof. Rudolf Kessler, Germany) 10:00 - 10:25 Hyperspectral imaging of food (Ferenc Firtha, Hungary) 10:25 - 10:45 Break 10:45 - 11:10 Hyperspectral imaging of biomaterials in agriculture and other fields (Philippe Vermeulen, Belgium) 11:10 - 11:35 Hyperspectral imaging of wood (Ingunn Burud, Norway) 11:35 - 12:00 Hyperspectral-/ Chemical Imaging as Key Technology in Sensor Based Sorting Applications (Matthias Kerschhaggel, EVK Graz, Austria) 12:00 - 12:25 Miniature multifunctional hyperspectral camera based on novel IMEC HSI technology (Manuel Cubero-Castan, Gamaya SA, Switzerland) 12:25 - 14:00 Lunch break & Poster Session 14:00 - 15:00 Demonstration of hyperspectral scanning procedure - hands on activities with different kinds of samples and different equipment (EVK integrated solution and Headwall Photonics VNIR) – you can bring your own samples! 15:00 - 18:00 Chemometry course (Jeremy Shaver, Eigenvector, USA) 18:00 - 18:15 Closing 18:15 - 20:00 Wine and cheese party


2

LECTURES


3

Hyperspectral Imaging – a general introduction

Rudolf W. Kessler Steinbeis Transferzentrum Prozesskontrolle und Datenanalyse, Herderstr. 47, 72762 Reutlingen, Germany, [email protected]

Intelligent manufacturing has attracted enormous interest in recent years. Optical spectroscopy will play a major role in the sensor technology as it provides simultaneously chemical (by absorption) and morphological (by scatter) information [1]. Special emphasis in the future will be given to measure not only the chemical entities but also their lateral distribution in an object. Spectral Imaging, or also called Chemical Imaging is an emerging field with applications ranging e.g. to find biomarkers in a tissue but also to control and qualify 100% of food. In the so-called whiskbroom imaging (“mapping”), the object is measured point-by-point with the advantage of high spectral and spatial resolution. A “staring” imager (“imaging”) takes two-dimensional images in a series, one after the other at several wavelengths. In pushbroom imaging (“line scanning”) the object is imaged along the y-axis using the linescan method and is recorded in full through the movement of the object in the y-direction. This is ideally suited for inline quality control in process analytical technology (PAT) [2]. The lecture will show, that the sensitivity and selectivity of each individual technology in combination with the used wavelength range has its limitations due to the structure of the measured specie and the used optical configuration. Furthermore, the suitability and robustness of each technology for inline applications is pre-determined by the selection of appropriate light illumination sources and the selected detectors. In any application, a key issue is to find the causal link between the measured spectral features, their spatial distribution in the x-, y- and possibly z-direction and the target quality. The paper will outline the concept of inline spectral imaging to characterize the chemical AND morphological features simultaneously and will show examples from the food and pharmaceutical industry, some biomedical applications and will present examples of applications in the manufacturing industry with a focus on the wood, paper and pulp industry.

[1] R. W. Kessler, Perspectives in process analysis. J. Chemometrics, 2013, 27: 369–378. doi: 10.1002/cem.2549 [2] B. Boldrini, W. Kessler, K. Rebner and R. W. Kessler, Hyperspectral imaging: a review of best practice, performance and pitfalls for inline and online applications, Journal of Near Infrared Spectroscopy 2012, 20, 438–508, doi: 10.1255/jnirs.1003


4

Hyperspectral imaging on Food - data acquisition and processing tools, some beneficial application field Ferenc Firtha Corvinus University of Budapest, Department of Physics and Control, Somlói út 14-16, H1118 Budapest, Hungary; [email protected] Although the hyperspectral method has several advantages, like remote sensing, segmentation of object and scanning non-homogeneous surface, it also has some disadvantages comparing to a NIR spectrometer. Non-isolated environment probably gives lower signal to noise ratio (S/N). After varying the field of view or illumination, the measurement must be recalibrated. Uneven surface necessarily results indefinite geometry (illumination/observation angles). „Argus” measurement control software will be introduced, that supports calibration, control data acquisition and ensures high S/N. The hypercube gained contains enormous amount of data (~50MB). A Matlab algorithm (CuBrowser) can help to reduce the data. Region of interest (ROI) areas can be selected manually and their average spectra, the 1st and 2nd derivatives are displayed. The effect of indefinite geometry can also be eliminated by normalizing spectra of pixels. The spectra and the derivatives of ROIs can be saved as samples for automatic segmentation (clustering) or as independent parameters of statistical analysis. Some well-known statistical model will be presented, like the Fisher’s Discriminant Analysis for dimensionality reduction and classification, and Partial Least Squares regression for predicting continuous dependent variables. Finally some major, promising application fields of hyperspecral method are shown: 1.

Invisible fungal infection of mushroom can be early detected since the spectra of spots differ from healthy areas. The non-homogeneity holds the information.

2.

Remote sensing can be useful on an industrial line but also capable to look through a film covering a product like cheese. The activity of some enzymes can be tracking during storage. In another experiment different kind of cheeses are classified on the base of their spectra, and also the optimal storing temperature is determined.

3.

The Advantage of a handicap: Since hypercube can be normalized by pixels, in case of uneven surface (tea leaves, coffee) the hyperspectral system can serve much higher S/N than a NIR spectrometer.

4.

When spatial distribution holds the information: In case of marzipan, the fructose produced by some enzymes cannot distinguished by spectra. Since fructose is hydrophil, the spatial distribution shape of moisture content can detect fructose content.


5

Detection of undesirable substances by NIR hyperspectral imaging spectroscopy and chemometrics in agriculture Philippe Vermeulen*, Juan Antonio Fernández Pierna, Pierre Dardenne, Vincent Baeten Walloon Agricultural Research Centre, Valorisation of Agricultural Products Department, Henseval' Building, Chaussée de Namur 24, 5030 Gembloux, Belgium; *[email protected] Until now, safety and quality control of agricultural products has been often carried out using classical physico-chemical methods that have limitations in terms of optimal implementation for the automatic control of products. Recent developments in analytical instrumentation and data processing methods have led to an increased use of vibrational spectroscopic techniques, as alternative methods replacing these physico-chemical techniques. The main aim of this work is to show that combining one of these recent analytical techniques, the NIR hyperspectral imaging system, with adequate chemometric tools can greatly improve the control of safety and quality of agricultural products. Two case-studies have been selected aiming to detect and quantify contamination in the cereals production control sector and for the assessment of resistant plants in sugar beet breeding programs. These studies used NIR hyperspectral imaging combined with relevant decision rules based on chemometric tools, spectral profiles and morphological information. The first study was dedicated to develop a complete procedure for detecting ergot bodies in cereals. The study sought to transfer this procedure from a pilot online NIR hyperspectral imaging system at laboratory level to a NIR hyperspectral imaging system at industrial level and to validate the latter. The topic of the study was extended to the detection of various impurities and contaminants in cereals. The second study focused in the contamination of plants by pathogens and showed the potential of this technology for detecting and quantifying cyst nematodes in sugar beet roots as well as for detecting and following the cercospora leaf spots development on sugar beet leaves. Acknowledgement: The research on ergot detection in cereals has received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° KBBE-211326 (CONffIDENCE). We are also grateful to Olivier Amand and Alain Tossens (SESVANDERHAVE, Belgium) for the collaboration on studies performed on sugar beet breeding program. [1] Vermeulen P., Fernández Pierna J.A., Van Egmond H., Zegers J., Dardenne P. & Baeten V. (2013). Validation and transferability study of a method based on near-infrared hyperspectral imaging for the detection and quantification of ergot bodies in cereals. Analytical and Bioanalytical Chemistry, 405: (24), 7765-7772 [2] Fernández Pierna J.A., Vermeulen P., Amand O., Tossens A., Dardenne P. & Baeten V. (2012). NIR hyperspectral imaging spectroscopy and chemometrics for the detection of undesirable substances in food and feed. Chemometrics and Intelligent Laboratory Systems, 117: 233-239


6

Hyperspectral imaging on wood Ingunn Burud1, Andreas Flo1, Anna Sandak2, Jakub Sandak2

1. Department of Mathematical Sciences and Technology, Norwegian University of Life Sciences, N-1432 Aas, Norway; [email protected] 2. Trees and Timber Institute IVALSA/CNR, via Biasi 75, 38010 San Michele all’Adige (TN), Italy Hyperspectral near infrared imaging is an increasingly used technique to study wood properties as a complementary tool to traditional point spectroscopy. The advantage of hyperspectral imaging is that a full spectrum is obtained in each pixel of the image. However, there are some challenges connected to studies of wood surfaces. Wood can often have a rough surface resulting in reflections from the light source for the imaging. Structures in the wood caused by e.g. growthrings, knots and sometimes cracks yield natural variations in colour and fungal growth on the surface. Moreover, wood substrates that are, or have been exposed in an outdoor environment will have a colour change due to photo degradation of lignin, wetting/leaching of the upper layer of the wood surface and growth of a variety of wood discolouring fungi. Several applications of hyperspectral imaging on wood will be presented, both from laboratory measurements with controlled light sources and field experiements in outdoor environment using natural sun light (Fig. 1 a and b). Data aquisition, preprocessing and various algorithms for data analysis will be discussed for these applications. Although most applications of hyperspectral imaging are performed in reflectance mode, a recent experiment performed in transmission mode will also be presented. This experiment was carried out on wood samples of thickness ~100 μm (Fig. 1c).

a b c Figure 1: a) Mould growth on wood surfaces observed by hyperspectral imaging in laboratory; b) Outdoor setup for hyperspectral imaging of wood; c) weathered thin samples measured in transmission mode with hyperspectral camera.


7

Hyperspectral-/ Chemical Imaging as Key Technology in Sensor Based Sorting Applications Matthias Kerschhaggl

EVK DI Kerschhaggl GmbH, Josef-Krainer-Strasse 35 | A-8074 Raaba/Graz, Austria; [email protected] Sensor based sorting is a major building block for modern production lines where input streams of raw materials are processed at high speeds (~ m/s). In this regard sensor heads such as RGB and near infrared (NIR) cameras, metal detectors, X-ray sensors and laser inspection are state of the art technologies used for the grading and sorting of feedstock. A very recent development in this field has been the introduction of so called hyperspectral imaging (HSI) cameras in the NIR regime. This technology allows the inference of entire spectra per imaging pixel in the wavelength range between 950-2300 nm at very high rates (~500 Hz for an entire image frame). Such it is possible to identify chemical constituents in bulk material along the same lines as done in the analytical laboratory e.g. using a standard spectrometer but spatially resolved across the field of view. Such it is possible not only to sort feedstock according to its chemical composition but also do quantitative spectroscopic imaging analysis in the production line. This talk will present several use cases from various industries such as mining, pharmaceutical, waste recycling and food production using the smart hyperspectral imaging camera HELIOS (EVK, Raaba/Austria) operating in the NIR regime (950-1700 nm).


8

Miniature multifunctional hyperspectral camera based on novel IMEC HSI technology

Manuel Cubero-Castan

Gamaya SA, Bâtiment C, EPFL Innovation Park, 1015 Lausanne, Switzerland; [email protected] Hyperspectral imaging constitutes a single most effective method of large-scale monitoring and analysis of vegetation with a proven capability in classification of plant species; early detection, diagnosis and control of plant diseases; timber quality assessment, stress detection and growth monitoring; as well as detection and control of invasive species. Scientifically proven benefits of airborne HSI monitoring of farmland include increased yield by 7-25 %, reduced use of chemicals by 40 %, decreased disease-related losses by up-to 70°%. Despite the many proven benefits, the adoption of the HSI methods in both research and commercial applications has been thus far impeded by the high cost of equipment and the excessive data acquisition and processing complexity. In particular the existing HSI satellites do not provide the spatial and the temporal resolution required for most applications. On the other hand, the existing airborne solutions are too expensive and operationally complex. Against this background, we are leverages the cutting edge advances in the domains of HSI technology, Unmanned Aerial Systems (UAS) and spectral analytics to develop and deploy a comprehensive solution for monitoring and spectral diagnostics of vegetation. The system will employ a new class of miniature HSI sensors developed by IMEC in collaboration with Gamaya and the Geodetic Engineering Laboratory of EPFL.


9

Hands-on Classification and Exploratory Analysis for Multivariate Images

Jeremy M. Shaver Eigenvector Research, 3905 West Eaglerock Dr., Wenatchee, WA, 98801, USA; [email protected] The use of multivariate analysis on images is becoming both standard and necessary as we begin to study increasingly complex samples and the power of the imaging instrumentation increases. Information-rich data presents both opportunities to improve specificity of models and challenges to avoid over-fitting and being limited by computational resources. This workshop will present a hands-on introduction to using multivariate analysis to examine images with a focus on classification exploratory analysis methods. Topics will include an introduction to general image-specific considerations for multivariate analysis, importing and basic manipulations, pre-processing, typical analysis methods, and interpreting image-based results. The workshop will have a significant hands-on focus so that users will be able to leave with a functional knowledge of multivariate image analysis (MIA.)


10

DEMOS


11

Demonstration of hyperspectral scanning procedure hands on activities with different kinds of samples I Matthias Kerschhaggl

EVK DI Kerschhaggl GmbH, Josef-Krainer-Strasse 35 | A-8074 Raaba/Graz, Austria; [email protected] This hands-on training will focus on sample inspection with the hyperspectral imaging (HSI) camera HELIOS (EVK, Raaba/Austria) taken from various industrial applications. Samples from the mining, pharmaceutical, waste recycling and food industries have been prepared for a live presentation of the sorting capabilities of this industrial near infrared HSI system (NIR 950-1700 nm). Apart from classification results the principle methodology of data acquisition and chemometric analysis will be shown.

Demonstration of hyperspectral scanning procedure hands on activities with different kinds of samples II Kevin Lynch

Headwall Photonics Inc., Fitchburg 01420, 601 River Street, USA; [email protected] This hands-on training will focus on sample inspection with a hyperspectral imaging (HSI) system of Headwall Photonics (VNIR, 400 – 1000 nm). The focus will be on the brief explanation of the principles of hyperspectral imaging followed by a demonstration of the process of gathering hyperspectral data from selected samples. Participants are invited to bring their own samples, and can analyze the data during the chemometry course offered at the workshop.


12

POSTER


13

Hyperspectral Root Imaging System Thomas Arnold1, Gernot Bodner2, Raimund Leitner1 1. Carinthian Tech Research AG, Molecular Imaging, Europastrasse 4/1, 9521 Villach, Austria; [email protected] 2. University of Natural Resources and Life Sciences, division of agronomy, KonradLorenz-Strasse 24, 3430 Tulln, Austria; [email protected]

The development of a hyper-spectral root imaging system for the acquisition of spatially resolved NIR spectroscopy data of rhizoboxes is presented. The system is used by the University of Natural Resources and Life Sciences Vienna (BOKU) for their research activities in the field of drought resistance of roots. In contrast to imaging using visible light (380 nm to 780 nm) hyper-spectral NIR imaging (1000 nm to 2500 nm) allows us to discriminate essential features for achieving the aims of the actual research project. The increased image contrast in the NIR range allows roots to be segmented from soil and additional information, e.g. basic root-architecture, to be extracted. In addition, the water absorption bands in the NIR wavelength range are used to measure the water content and to estimate the age of the roots. The poster describes the hardware setup of the hyper-spectral root imaging system and the adopted algorithms. Moreover, preliminary results of soil water content estimation are presented as well as an outlook of future research activities.


14

NIR hyperspectral imaging as a tool to map the biochar content in a soil profile Ingunn Burud1, Christhophe Moni2, Andreas Flo1, Cecilia Futsaether1, Markus Steffens3, Daniel Rasse2

1. Department of Mathematical Sciences and Technology, Norwegian University of Life Sciences, N-1432 Aas, Norway 2. Bioforsk – Norwegian Institute for Agricultural and Environmental Research. Frederick A. Dahls vei 20, Aas, Norway 3. Lehrstuhl für Bodenkunde, Department für Ökologie und Ökosystemmanagement, Wissenschaftszentrum Weihenstephan für Ernährung, Landnutzung und Umwelt, Technische Universität München, 85350 Freising-Weihenstephan, Germany

Biochar is a promising technology for sequestering C in soils in the form of charcoal obtained from the pyrolysis of biomass. In addition, biochar generally increases soil fertility due to improvements in soil water content, pH and nutrient retention. The implementation of C credit schemes calls for cheap and reliable systems for quantification and mapping of biochar products in soil. In addition, many research questions are linked to the size and distribution of biochar particles in soils. In this work NIR (Near infrared) hyperspectral reflectance imaging has been carried out with the aim of identifying biochar particles and mapping biochar concentrations in a soil core. Analyses of the hyperspectral images have been carried out on median spectra from hyperspectral images of ground soil samples, and on the pixel level on soil profiles from the field. Results from PLS regression and Spectral Angle Mapper (SAM) analyses conducted on the hyperspectral images were compared with distribution of biochar concentrations measured by combined total C and 13C signature analyses. The results indicate that biochar can be identified in soil profiles using hyperspectral NIR images. Quantifying the amount of biochar, however, is challenging on the pixel level due to the rough surface of the soil profile, non-homogenous distribution of the biochar and varying moisture conditions.


15

Spectral signatures as a potential tool for the discrimination of fungal infections on grapevine (Vitis vinifera) Daniel Molitor1,2, Marie Theres Khuen1, Christa Schefbeck1,2, Franz Kai Ronellenfitsch2, Martin Schlerf2, Marco Beyer2, Katharina SchödlHummel1, Ulrike Anhalt1, Astrid Forneck1 1. University of Natural Resources and Life Sciences, Department of Crop Sciences, Division of Viticulture and Pomology, Konrad-Lorenz-Strasse 24, A-3430 Tulln, Austria; [email protected] 2. Luxembourg Institute of Science and Technology, Environmental Research and Innovation (ERIN) Department, 41, rue du Brill, L-4422 Belvaux, Luxembourg; [email protected] The fungal pathogens Botrytis cinerea and Penicillium expansum are causing economic damages on grapevine worldwide. Especially the simultaneous occurrence of both often results in off-flavours diminishing wine quality. For the classification of grape quality (linked to pay-out prices for grape producers) as well as for the determination of targeted enological treatments, the knowledge of the level of fungal attack is of highest interest. However, visual assessment and pathogen discrimination are challenging and cost-intensive. Consequently, a pilot laboratory study aimed at (i) detecting differences in spectral signatures between grape berry lots with different levels of infected berries (B. cinerea and/or P. expansum) and (ii) detecting links between spectral signatures and biochemical as well as quantitative molecular markers for fungal attack. To this end, defined percentages (infection levels) of table grape berries were inoculated with fungal spore suspensions. Spectral measurements were taken using a FieldSpec 3 Max spectroradiometer (ASD Inc., Boulder/Colorado, USA) in regular intervals post-inoculation. In addition, fungal attack was determined enzymatically (enzymatic assay and photometric measurement of gluconic acid concentrations) and quantitatively (real-time PCR). Preliminary results indicate that based on spectral signatures, a discrimination of P. expansum and B. cinerea infections as well as of different B. cinerea infection levels is possible. Real-time PCR analyses, detecting DNA levels of both fungi, showed yet a low detection level. Consequently, present assays need further improvements. Whereas the gluconic acid concentrations turned out to be specific for the two fungi tested (B. cinerea vs. P. expansum) and thus may serve as a differentiating biochemical marker. Correlation analyses between spectral measurements and biological data (gluconic acid concentrations, fungi DNA) as well as further common field and laboratory trials are targeted.


16

Investigation of fungal infections on white button mushroom caps using hyperspectral imaging technique Viktória Parrag1, József Felföldi1, András Geösel2, László Baranyai1, Dániel Szöllősi1, Ferenc Firtha1 1. 2.

Corvinus University of Budapest, Department of Physics and Control, Somloi út 14-16, H-1118 Budapest, Hungary; [email protected] Corvinus University of Budapest, Department of Vegetable and Mushroom Growing, Ménesi street 44, 1118 Budapest, Hungary

In the last decades microscopic fungi caused great losses for the mushroom industry. The most aggressive infections of white button mushroom are cobweb disease and green mold. These infections can spread very rapidly before and also after the harvesting. Hyperspectral imaging is a promising method to investigate these fungal infections. It can provide both spectral and spatial information from the measured object, so the development of the fungi can be followed. Different experiments were carried out to investigate these infections. Mushroom samples were photographed and the hyperspectral images were acquired in the wavelength range of 900-1700 nm. The image processing system and the sensor were controlled by Argus hyperspectral software [1]. Regions of interests were selected manually on the hyperspectral images using CuBrowser Matlab algorithm [2] and the average spectra of areas were saved. On the spectra a simple normalization method and a Savitzky-Golay smoothing was carried out. In one experiment samples were divided and the first part was infected with Cladobotryum dendroides (main causal agent of cobweb disease). The classification was carried out using Support Vector Machine method and the validation was carried out with Monte Carlo crossvalidation. According to our results there are relevant spectral differences between the samples at 1450 nm and around 1080 nm. 90.91% of the samples were correctly classified with SVM method and 88.9% of the spectra were correctly classified during the validation. In the next experiment samples were divided into 3 groups: infected manually with Cladobortyum, infected with Trichoderma (green mold) and control (treated with distilled water). The samples were stored chilled, under controlled conditions for 5 days, during this period symptoms were not visually perceptible. The classification was carried out using principal component analysis (PCA) and linear discriminant analysis (LDA). The discrimination using LDA between infected and control samples was remarkably successful and this method was also able to separate the two types of infection. [1] F. Firtha, Argus, ftp://fizika2.kee.hu/ffirtha/Argus-CuBrowser.pdf, (2010), last accessed 15.12.2014 [2] F. Firtha, CuBrowser ftp://fizika2.kee.hu/ffirtha/Argus-CuBrowser.pdf, (2012), last accessed 15.12.2014


17

Multivariate analysis of hyperspectral data generated by Raman Imaging on differentiating xylem of Spruce Batirtze Prats-Mateu1, Notburga Gierlinger1 1. University of Natural Resources and Life Sciences, Department of Material Sciences and Process Engineering, Institute of Physics and Material Sciences, Peter-Jordan-Strasse 82, 1190 Vienna, Austria; b.prats‐[email protected], [email protected] Confocal Raman Microscopy is a powerful chemical analytical technique with a high spatial resolution (300nm) allowing spectral (chemical) mapping in context with the microstructure (xy direction). This means, thousands of spectra are generated when analysing few square micrometres. This technique has been used for revealing the spatial distribution of the wood components by integrating the characteristic single bands of the functional groups of interest (univariate approach) [1-3]. However the univariate method had limitations in elucidating single molecule variability and to overcome the generated large amount of data. Since plant cell wall spectra are quite complicated and a mixture of overlapping bands, different multivariate approaches e.g. “Non-linear matrix factorization” (NMF) or “Vertex Component Analysis” (VCA) have also been developed to unmix underlying bands and offer a fast method to unravel the different wood polymers as well as variations at the molecular level with subpixel resolution [4-6]. Here we present two multivariate analysis approaches to follow in situ lignification in wood: NMF (iterative) and VCA (non-iterative) for the analysis of hyperspectral data generated by non-destructive Raman Imaging on the differentiating area of the stem of a 7-years old Spruce (Picea abies) [7]. Acknowledgement: This research was supported by the Austrian Science Fund (FWF): START Project [Y-728-B16]. [1] Agarwal, U. P. (2006). Raman imaging to investigate ultrastructure and composition of plant cell walls: distribution of lignin and cellulose in black spruce wood (Picea mariana). Planta 224(5): 11411153. [2] Gierlinger, N.,Schwanninger, M. (2006). Chemical imaging of poplar wood cell walls by confocal Raman microscopy. Plant Physiol 140(4): 1246-1254. [3] Gierlinger, N.,Schwanninger, M. (2007). The potential of Raman microscopy and Raman imaging in plant research. Spectrosc-Int J 21(2): 69-89. [4] Gierlinger, N., Keplinger, T.,Harrington, M. (2012). Imaging of plant cell walls by confocal Raman microscopy. Nat Protoc 7(9): 1694-1708. [5] Gierlinger, N. (2014). Revealing changes in molecular composition of plant cell walls on the micron-level by Raman mapping and vertex component analysis (VCA). Front Plant Sci 5. [6] Geladi, P., Grahn, H.,Manley, M. (2010). Data Analysis and Chemometrics for Hyperspectral Imaging. Raman, Infrared, and Near-Infrared Chemical Imaging, John Wiley & Sons, Inc.: 93-107. [7] Prats-Mateu, B., Stefke, B., Gierlinger, N. (2015). Following lignification in Native Cell Walls by Raman Confocal Microscopy (unpublished data).


18

Hyperspectral Imaging for Food Quality Analysis

Jinchang Ren, Timothy Kelman, Tong Qiao, Paul Murray, Stephen Marshall Hyperspectral Imaging Centre, University of Strathclyde, Department of Electronic and Electrical Engineering, Glasgow, G1 1XW, U.K.; [email protected] As a key area which requires reliable, nondestructive and accurate measures, food quality analysis is one important application of hyperspectral imaging (HSI). In this talk, the activities and advanced data analytics in Strathclyde Hyperspectral Imaging Centre is summarised. Followed by a discussion of some state-of-the-art data processing techniques, four different applications of HSI for food quality analysis are discussed. These include shelf life estimation of baked sponges; beef quality prediction; classification of Chinese tea leaves; and classification of rice grains. The first two focus on using HSI for regression based quality related data prediction, and the final two are for data classification. The estimated objective measurements can then be used for grading and assessment of the food quality. Comprehensive experimental results will be reported to demonstrate the efficacy of our proposed data analysis techniques and the great potential of HSI.

Fig. 1: VNIR HSI System

Fig. 2: HSI of Baked Sponges (a) Spectral profile ageing chocolate sponge and (b) HSI scores produced from reflectance at 970 nm compared with organolpetic data for chocolate sponge.

Acknowledgement: Thanks to the support from Quality Meat Scotland, Lightbody of Hamilton, Finsbury Food Group, Scottish Funding Council, University of Strathclyde and other partners. [1] J. Ren, et al., “Effective feature extraction and data reduction with hyperspectral imaging in remote sensing,” IEEE Signal Processing Magazine, 2014 [2] J. Zabalza, et al., “Novel folded-PCA for improved feature extraction and data reduction with hyperspectral imaging and SAR in remote sensing,” ISPRS Journal of Photogrammetry and Remote Sensing, 2014 [3] J. Zabalza, et al., “Novel two dimensional singular spectrum analysis for effective feature extraction and data classification in hyperspectral imaging,” IEEE Transactions on Geoscience and Remote Sensing, 2015 [4] T. Qiao, et al., “Quantitative prediction of beef quality using visible and NIR spectroscopy with large data samples under industry conditions,” Journal of Applied Spectroscopy, 2015. [5] T. Kelman et al, “Effective classification of Chinese tea samples in hyperspectral imaging,” Artificial Intelligence Research, 2013 Workshop on Hyperspectral Imaging | BOKU‐UFT

19

Quick assessment of Collagen preservation in fossil bones using Hyperspectral Imaging Damien Vincke1, Rebecca Miller², Édith Stassart2, Marcel Otte2, Pierre Dardenne1, Matthew Collins3, Keith Wilkinson4, John Stewart5, Vincent Baeten1, Juan Antonio Fernández Pierna1 1. Walloon Agricultural Research Centre, Valorisation of Agricultural Products Department, Henseval' Building, Chaussée de Namur 24, 5030 Gembloux, Belgium, [email protected], [email protected], [email protected], [email protected] 2. University of Liege, Service of Prehistory, Place du XX Août 7, Bâtiment A1, 4000 Liège, Belgium, [email protected], [email protected], [email protected] 3. University of York, Department of Archaeology, Wentworth Way, York, North Yorkshire YO10 5DD, United Kingdom, [email protected] 4. University of Winchester, Department of Archaeology, Winchester SO22 4NR, United Kingdom, [email protected] 5. Bournemouth University, Faculty of Science and Technology, Talbot Campus, Fern Barrow Poole, Dorset BH12 5BB, United Kingdom, [email protected] Collagen is a critical material in archaeology required for different analyses (radio carbon dating, ancient DNA, etc.). For such analyses, actually, archaeologists are faced with the issues of cost and time, and the risk of failure if collagen preservation is insufficient. Rapid and non-destructive techniques are needed to screen, at laboratory and on-site, a large number of bones to detect and quantify the amount of collagen preserved. Vibrational spectroscopy techniques such as Near Infrared (NIR) and NIR hyperspectral imaging fulfil all these conditions. The results obtained in this work indicate that NIR hyperspectral imaging combined with classic chemometric tools enables the detection of specific spectral bands characteristic of collagen as well as the analysis of the collagen distribution (taphonomy) within and between different strata [1].

Acknowledgement: The ArcheoNIR Project is funded by the Fonds De La Recherche ScientifiqueFNRS,Fonds de la Recherche Fondamentale Collective (FRFC), Project number F.FRFC2.4621.12. The Trou Al'Wesse Project, directed by Rebecca Miller and Marcel Otte (University of Liège, Service of Prehistory, Belgium), issupported by annual subsidies (02/16341 to 13/19227) granted by the Service Public de Wallonie (SPW). We are also grateful to Ouissam Abbas (CRA-W) and the technical staff of all of the teams. [1] Vincke D., Miller R., Stassart E., Otte M., Dardenne P., Collins M., Wilkinson K., Stewart J., Baeten V., Fernández Pierna J.A., 2014, “Analysis of collagen preservation in bones recovered in archaeological contexts using NIR Hyperspectral Imaging”, Talanta, 125, 181-188. doi:10.1016/j.talanta.2014.02.044


20

Inspection of log quality by hyperspectral imaging A. Zitek1, F. Firtha2, K. Böhm1, V. Parrag2, J. Sandak3, B. Hinterstoisser1 1. Institute of Wood Technology and Renewable Materials, Department of Material Sciences and Process Engineering, University of Natural Resources and Life Sciences Vienna BOKU, Peter Jordan Str.82, A-1190 Vienna, Austria; [email protected] 2. Physics-Control Department, Faculty of Food Science, Corvinus University of Budapest, Budapest Somlói út 14-16, H-1118, Hungary 3. Trees and Timber Institute/National Research Council, IVALSA/CNR, Via Biasi 75, San Michele All’Adige, Italy The accurate assessment of wood quality and detection of any wood deficiencies at an early stage of the forest-wood processing chain is a highly relevant step for a sustainable organization of the following transformation stream. Hyperspectral imaging (HSI) has been recognized as a powerful technique for inspecting organic matter, including wood, and may provide usable info regarding wood quality in a spatially resolved manner. A wooden disc of a ~80 year old Norway spruce (Picea abies (L.) Karst) was investigated in this pilot study by both, hyperspectral imaging and FT-NIR spectroscopy. The surface of the disk included portions of rough structure, and fairly visible fungal infestation. The sample was evaluated in two moisture conditions including dry and partly wet conditions. It was found that wooden disk areas affected by fungi and/or structural abnormalities could be clearly identified on the dry and wet wood surface by means of hyperspectral imaging at selected wavelengths or by linear wavelength processing. The influence of wood surface roughness was also negligible, indicating the potential of this technique for application on the freshly harvested logs. Analysis of FT-NIR spectra revealed further details on the woody polymers degradation by fungi. The comparison of the results gathered by HSI and FT-NIR technique showed, that FT-NIR - as a common method to determine wood quality - can be effectively used as a reference method and for calibration of the wood quality grading system. Summarizing, hyperspectral imaging proved to be a promising technique that, when combined with adequate chemometric models and image processing, could significantly support the on-site quality grading of logs. Several issues have to be addressed in the subsequent studies, including adjustments due to the uneven moisture distribution, excessively rough surfaces, detection of cracks, and identification of soil or machine oil contaminations and different lightning conditions. The combination of spectroscopic analysis with digital image processing techniques will be further explored to improve the system’s capability for detection and identification of most relevant wood defects. Acknowledgement: We would like to dedicate this work to the memory of Dr. Manfred Schwanninger (1963-2013). This research has been conducted within frame of the SLOPE project (FP7Collaborative Project – 604129) “Integrated proceSsing and controL systems fOr sustainable forest Production in mountain arEas”. [1] F. Firtha, Argus hyperspectral acquisition software, ftp://fizika2.kee.hu/ffirtha/ArgusCuBrowser.pdf, (2010), accessed 20.11.2014 [2] F. Firtha, CuBrowser hyperspectral data processing algorithm ftp://fizika2.kee.hu/ffirtha/ArgusCuBrowser.pdf, (2012), accessed 20.11.2014


21