hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

13
Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process Silvia Serranti Andrea Fabbri Giuseppe Bonifazi Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Upload: silvia

Post on 19-Dec-2016

220 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

Hyperspectral imaging techniquesapplied to the monitoring of wine wasteanaerobic digestion process

Silvia SerrantiAndrea FabbriGiuseppe Bonifazi

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 2: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

Hyperspectral imaging techniques applied to themonitoring of wine waste anaerobic digestion process

Silvia SerrantiAndrea FabbriGiuseppe BonifaziSapienza-Università di RomaDipartimento di Ingegneria Chimica Materiali

AmbienteVia Eudossiana 18, 00184 Rome, ItalyE-mail: [email protected]

Abstract. An anaerobic digestion process, finalized to biogas production,is characterized by different steps involving the variation of some chemicaland physical parameters related to the presence of specific biomasses as:pH, chemical oxygen demand (COD), volatile solids, nitrate (NO−

3 ) andphosphate (PO−

3 ). A correct process characterization requires a periodicalsampling of the organic mixture in the reactor and a further analysis of thesamples by traditional chemical-physical methods. Such an approach isdiscontinuous, time-consuming and expensive. A new analytical approachbased on hyperspectral imaging in the NIR field (1000 to 1700 nm) isinvestigated and critically evaluated, with reference to the monitoring ofwine waste anaerobic digestion process. The application of the proposedtechnique was addressed to identify and demonstrate the correlation exist-ing, in terms of quality and reliability of the results, between “classical”chemical-physical parameters and spectral features of the digestatesamples. Good results were obtained, ranging from a R2 ¼ 0.68 and aRMSECV ¼ 12.83 mg∕l for nitrate to a R2 ¼ 0.90 and a RMSECV ¼5495.16 mgO2∕l for COD. The proposed approach seems very usefulin setting up innovative control strategies allowing for full, continuouscontrol of the anaerobic digestion process. © 2012 Society of Photo-OpticalInstrumentation Engineers (SPIE). [DOI: 10.1117/1.OE.51.11.111708]

Subject terms: hyperspectral imaging; partial least squares; anaerobic digestion;biogas; process monitoring; wine waste.

Paper 120341SS receivedMar. 8, 2012; revisedmanuscript receivedMay 15, 2012;accepted for publication May 16, 2012; published online Jun. 14, 2012.

1 IntroductionImproved waste production and the corresponding increasein energy consumption has forced the scientific communityin the past few years to find solutions to solve both problems.Regarding agro-industrial by-products, biological and ther-mal treatments have been investigated and evaluated fortheir valorization and to reduce the yields brought to thelandfills. Composting, incineration, pyrolysis, and anaerobicdigestion (AD) are the most common technologies employedin organic waste treatment. Among the possible availabletechniques, AD represents a perfect “green solution.” Treat-ing organic wastes by AD presents two major advantages:the production of biofuel (through methane production) tobe utilized in combined heat and power (CHP) plants for pro-duction of thermal and electrical energy; the elimination ofwaste by the recovery of material commonly sent to landfillsand the utilization of digestate as fertilizer.

AD is carried out in many naturally occurring anaerobicenvironments, including water courses, sediments, water-logged soils and the mammalian gut. It occurs in severalsteps, starting with the fermentation process (hydrolysis andacidogenesis) of all complex substances (e.g., fats, proteins,carbohydrates) present in organic matter that are convertedinto monomers and then into volatile fatty acid (VFA). Inacetocenesis the transformation of VFA and alcohol intoacetic acid, hydrogen and carbon dioxide by acetogenicbacteria occurs. Finally, the methanogenesis methane is

produced by transforming acetic acid (acetoclastic methano-gens) and hydrogen and carbon dioxide (hydrogenophilicmethanogens). The chemical oxygen demand (COD) ofthe influent is reduced and constitutes a cheaper alternativefrom an expended energy point of view, because it producesa biogas-containing methane that could eventually coverenergy needs.1,2

The production of biogas through anaerobic digestionoffers several advantages3 in comparison with other wasteprocessing strategies, such as:

• a lower production of biomass in comparison toaerobic-based processing,4

• a better processing of wet wastes characterized by alow (less than 40%) dry solids content,5

• an efficient pathogen removal,6,7

• an absence of odor, as practically all the volatile com-pounds oxidatively decomposed upon combustion,8

• a compliance with many national waste strategiesimplemented to reduce the amount of biodegradablewaste entering landfills,9

• the possibility to utilize the digestate (resulting slurry)as fertilizer,10

• reduction of fossil fuels utilization for energyproduction.11

The degradation, conversion and stabilization of organicmaterials by a biological process in absence of oxygenproduces a gas mixture of CH4 and CO2 as main0091-3286/2012/$25.00 © 2012 SPIE

Optical Engineering 111708-1 November 2012/Vol. 51(11)

Optical Engineering 51(11), 111708 (November 2012)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 3: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

components, and H2, CO, H2S as minor components12 alongwith a microbial biomass.13

AD can be applied not only to agro-industrial byproductsbut also to the putrescible matter (e.g., kitchen, fruit andvegetable waste, green waste) contained in the urbanwaste. Such a fraction is usually about 32%. As a conse-quence, the high quantities of organic fraction present inurban, agricultural and industrial waste represent a bigresource to be utilized in anaerobic treatment.14 These lattermaterials are constituted by a heterogeneous matrix that isusually processed in order to select the best waste productpresenting constant characteristics and a good amount ofmethane production.

Many studies have been carried out in order to identifyand quantitatively assess the relationships existing betweenanimal-by-product (ABP), digestibility and the chemicalparameters of fresh matrices inside the digestor and theresulting digested bioslurries.15–19 In particular it is importantto monitor some specific chemical and physical parametersof the blend in the reactor, as nitrates20–23 are responsible forinhibitory effects on the kinetics of bio-methane productionand/or other parameters as COD, pH and volatile solids (VS).A full monitoring of the previously mentioned parameters,including process kinetics and detecting the presence ofinhibitory substances and/or phenomena related to possiblevariations of the physical-chemical parameters, can becarried out following a classical chemical analyses-basedapproach. However, it costs both money and time andfurthermore, requires specialized equipment operated bytrained personnel.

Classical near-infrared spectroscopy (NIRS) has been fre-quently reported as an easy-to-apply and low-cost analyticalapproach that can be profitably utilized to substitute many“classical” laboratory analyses to allow faster measurements.NIRS is based on the detection-and-analysis-effect of theinteraction between photons (1000 to 1700 nm) and thematerial/product under investigations. NIRS analyses areusually carried out in two steps, starting with a preliminarycalibration model and followed by a training procedure con-sisting of a calibration set that is created and used to find therelationship between the spectra and the parameters object ofthe investigation. The model is then tested on a test setthat does not include the values used during the calibrationphase. Following this strategy both quantitative (e.g., pro-ducts of interest concentrations, level of any property)and/or qualitative (e.g., belonging to a class) informationcan be predicted. For these reasons NIR spectroscopy rapidlybecame a well-consolidated technology utilized in manyresearch and industrial fields (e.g., food, agricultural, chemi-cal, pharmaceuticals, biological sectors, etc.).24–27 Further-more, the already mentioned “simple utilization,” andnon-destructive characteristics together with the low costsassociated with the analyses in respect to the accuracy ofthe results, determined its rapid growth and utilization forthe control of many processes, including anaerobic diges-tion.28–30 Following the same methodological approach, aninnovative spectral technique based on the acquisition ofhyperspectral images in the NIR range (1000 to 1700 nm)was fully investigated in this paper, with reference to theproduct’s evolution during an anaerobic digestion process.

In the past few years HSI has rapidly emerged as a fast,reliable, and low-cost quality control approach that does not

require any particular product sample preparation in manysectors, including food inspection,31–33 the pharmaceuticalindustry,34,35 medicine,36,37 artworks38 and polymersscience.39 Studies have been also carried out in solidwaste sectors: waste paper recovery,40 glass recycling,41

fluff characterization from ASR,42 bottom ash from munici-pal solid waste incinerators,43 compost products qualitycontrol,44 polymers45 and polyolefins 46,47 identification.

The aim of this paper is to present and evaluate the pos-sibility of using his-based techniques in the NIR field inorder to find a faster way to replace a traditional chemical,classical-spectroscopy-based technique and a physical analy-sis of the digestate during its evolution in anaerobic condi-tion and to utilize the HSI-NIR approach to perform anon-line monitoring of the chemical and physical parametersaffecting the digestion kinetic of a specific agro-industrialby-product: the grape marc (GM), which is the waste result-ing from wine production. After a full description of theinvestigated anaerobic process adopting a classical analyticalapproach, the HSI-NIR techniques and the different analyti-cal steps were presented with reference to digestate sampleimages. Such a strategy mimics a possible digestate multi-sampling-continuous-approach finalized to evaluate possiblevariations in digestate physical-chemical characteristicsusually not easily evaluated through a classical-single-sample-NIRS-based analytical technique. Digestate hyper-spectral data collection, in fact, allows acquiring not onlyinformation related to the chemical attributes of the sample,but also to its physical characteristics (e.g., size class char-acteristics and content of the solid fraction, solid/liquid ratio,etc.). Following a classical chemometric-based approach acalibration model was built and validated. The number ofinternal latent variables was optimized according to the low-est root-mean square error of calibration (RMSEC). In thisway the model is correctly projected and can predict a qua-litative or quantitative value starting from the NIR spectraof the sample.30 Results showed the proposed approach isparticularly feasible to be applied with reference to thebiomasses object of the study.

2 Wine Waste CaseThe wine sector represents a big source of organic wasteresulting both directly from wine production (e.g., skin,stalks and seeds) and/or from related maintenance/cleanoperations (e.g., fluids resulting from the vats and machinerycleaning). This agro-industrial compartment is very prosper-ous in Europe and South America, and is growing more andmore in North America, North Africa and Asia.

The International Organization of Vine and Wine (OIV)have redacted a world vitis-wine-cultural statistics report,where world grape productions are presented: 665.219t∕year of grapes were cultivated in 2007 which were treatedto produce about 265.994 hl of wine.48

The disposal of GM has long been a problem for wine-ries.49 GM typically is characterized by a moisture content ofabout 65% and represents as much as 20% of the wet weightof the original fruit. Once the juice has been extracted, theskin, stalks and seeds are all redundant. If not treated, themarc can produce a number of environmental hazardsranging from surface and ground water pollution to foulodors. Piles of GM attract flies and pests and can easilyspread diseases. Leachates (solutions of tannins and other

Optical Engineering 111708-2 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 4: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

compounds from the marc flowing out of material) can causeoxygen depletion in the soil and infiltrate surface, soil andground waters.

Historically, GM was used in distilleries to produce alco-hol50 or compost. In the last few years the oversupply ofgrape spirit led to a global drop in prices, and consequentlyproducers were no longer able to recover costs from sellingtheir GM to distilleries and had to pay for waste disposal.With increased wine production, it became imperative torelieve the oversupply of GM. The conversion or eliminationof large quantities of GM, as well as the other wine-making-related-by-products, represents a serious problem to solve,both in ecological and economic terms.51 AD can thus repre-sent an ecological and economic solution for GM valoriza-tion. Adopting this technology makes it possible to producean energy resource to be utilized as bio-fuel in combined heatand power (CHP) plants for electricity and heat production.Furthermore, it is also possible to utilize the digestate inagriculture as fertilizer.52

The utilization of the AD technology presents moreadvantages with respect to the composting process: bothsystems lead to a final, ready-to-use product in agriculture,but AD is an energy-making system, whereas composting isan energy consuming system, as the heat produced by com-posting is not recoverable. During the composting processparticular care has to be taken regarding organics decompo-sition: an uncontrolled decomposition of organic solid wastecan contaminate soil, water and air.53 On the other hand, dur-ing AD an unsuitable degradation process can produce adigestate that cannot be used as fertilizer because of thehigh concentration of nutrients such as nitrite (NO−

2 ) andnitrate (NO−

3 ). Nitrate, in fact, is denitrified in the firststage of decomposition; such a phase is important for metha-nation.21 A nitrate content (NO3 − N>50 mg kg−1 has aninhibitory effect on gas production. Furthermore it contri-butes to the production of a more negative redox-potential,a decrease of overall gas quality linked to the higher nitrogencontent of the biogas and carbon consumption, making itunavailable for methanation. Finally, a reduction of biogasproduction can also occur due to the high accumulation ofVFA that produces “negative” low pH values.

3 Materials and Methods

3.1 Waste (Substrate) and Inoculum Originand Collection

Tests have been carried out utilizing different GM: white(WGM) and red (RGM).

WGM (Fig. 1) were collected at a domestic wine producer(located in borgo Santa Maria, Latina, Italy). They werestored in tanks for 15 days after wine extraction. WGMwere constituted by different kinds of waste white grapes:Malvasia, Moscato Bianco and Trebbiano. White grapeswere treated before the wine extraction process with theremoval of the grape stalks. Such a process allowed thereduction of the presence of ligno-cellulosic products thatcan inhibit biogas production during the digesting phase.WGM were not fermented after the juice extraction process.A mixture of WGM, as collected, was utilized to perform thetests: the relative quantities of the constituting white grapeswere about equal to one-third each.

Fig. 1 Grape marcs (GM), e.g., waste resulting from wine production,utilized in the study. (a) White GM (WGM) and (b) red GM (RGM).

Fig. 2 Anaerobic pilot batch reactor architecture.

Optical Engineering 111708-3 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 5: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

RGM (Fig. 2) were provided by a Casale del Giglio, B.goLe Ferriere, Latina, Italy winery. The RGM storage was thesame as for the WGM. Red grapes were also treated beforethe juice extraction process adopting the same grape-stalksremoval process utilized for WGM. However, the biggestdifference from the WGM was that a fermentation processoccurred. Tests were carried out on a RGM sample represen-tative of the whole.

Cow manure (CM) was utilized as inoculum to activatethe process. Fresh CM was collected at from a local farmin the same area where the GM were collected. Particularcare was followed for manure selection: only CM withoutdetected presence of antibiotics was collected. Antibiotics,in fact, kill the methanogen bacteria responsible for biogasproduction. The characteristics of the substrates are reportedin Table 1.

3.2 Anaerobic Reactor Set-Up

To realize the anaerobic digester (Fig. 2) a galvanized-steelcylindrical bioreactor 50 cm in diameter and 112 cm inheight was built from a specifically modified commercial

boiler. The nominal and working volume of the bioreactorare 200 and 150 l, respectively, with pH and temperature(PT100) probes installed. The sensing units were positionedmid high on the reactor. A special gas caption unit was alsoinstalled inside the reactor to collect and analyze the biogasproduced into the gas line. The substrate in the reactor wasstirred by a mixing-unit operated at 60 rpm. Stirring was car-ried out 15 min every hour. The heating system operated byan interspace of 10 mm wrapping the cylindrical body of thereactor. The water was heated by electrical resistance tomaintain a constant temperature (35°C� 2°C) inside thereactor. A pump was utilized to circulate the water insidethe heating circuit. Biogas, after a dehumidification stage,was convoyed in a gasholder and quantitatively evaluated.

3.3 Classical Digestate Sampling and Analysis

At the beginning of the experiment, the following parameterswere measured to realize a full characterization of the raworganic matter and inoculum: total solids (TS), humidity, VS,ashes and COD (Table 1). During the fermentation tests,in addition to the parameters previously reported, nitrates,temperature and pH were also measured. All the analyses,including any fractioning and/or sub-sampling, were carriedout on a digestate sample as extracted from the anaerobicreactor. Digestate TS, humidity, VS, ashes, COD, andnitrates measurements were carried out by classical analyti-cal methods starting from samples collected from thedigestor by an off-line batch sampling every two days. Tem-perature and pH were continuously measured on-line andlogged in Microsoft Excel™. The user interface createdfor the collection and handling of the above-mentioned para-meters is reported in Fig. 3.

TS, Humidity, VS and ashes weight values were deter-mined according to ANPA (2001) standard methods; theCOD concentration values in liquid organic matter was mea-sured according to analytical methods illustrated byISRA-CNR54 and according to the Italian directive D.M.13/09/1999 for the COD concentration values in solid

Table 1 Characteristics of the substrate utilized to perform the tests.

Parameter WGM RGM CM

Humidity (%WW) 58.94 62.53 84.74

Total solid (%WW) 41.06 37.47 15.26

Volatile solid (%TS) 94.70 92.54 79.17

Ash (%TS) 5.30 7.46 20.83

COD (mgO2∕gGM and mgO2∕lIN) 172,370 120,660 –

WGM, white grape marc; WGM, red grape marc; CM, cow manure;WW, wet weight; TS, total solid.

Fig. 3 User-interface-developed adoption of the National Instruments’s Labview™ environment allowing for the collection and handling of all theprocess parameters affecting the mesophilic anaerobic digestion.

Optical Engineering 111708-4 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 6: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

organic matter. Nitrates were measured by ion chromatogra-phy utilizing a Shimazu CDD-10AVP™ analytical unit.Each sample was previously centrifuged (5 min at 3500 rpm)and the supernatant filtrated through 0.45 μm cellulosefilters. Details of the analytical procedure adopted can befound in Alvarez et al.55 Temperature was measured by aPT100 probe, connected to an ampere converter that gener-ates an analog signal (4 to 20 mA) acquired by an acquisitionmodule (National Instrument™ DAQ mod. NI9203). ThepH values were measured by a pH collecting device(J DIGITAL™ pH) installed in the anaerobic reactor. Asfor the temperature probe, an analog signal (4 to 20 mA) wasgenerated and then acquired by the same NI DAQ module.

3.4 HSI-NIR Analysis

An HSI-NIR-based architecture was utilized (Fig. 4). Thearchitecture was constituted by a NIR Spectral Camera™(Specim, Finland), embedding an ImSpector N17E™imaging spectrograph working in the spectral range from1000 to 1700 nm, with a spectral sampling/pixel of 2.6 nm,coupled with an Te-cooled InGaAs photodiode array sensor(320 × 240 pixels) with pixel resolution of 12 bits. A dif-fused light cylinder source, providing the required energyfor the sensing unit, was set up. The aluminum internal-coated cylinder embeds five halogen lamps producing a con-tinuous spectrum signal optimized for spectra acquisitionin the NIR wavelength range. The device works as a push-broom-type line scan camera providing full, contiguousspectral information for each pixel in the line. The trans-mission diffraction grating and optics provide high lightthrough-put and a high quality, distortion-less image forthe device. The result is constituted of a digital imagewhere each column represents the discrete spectrum valuesof the corresponding element of the sensitive linear array.The device is fully controlled by a PC unit equipped withthe Spectral Scanner™ v.2.3 acquisition/pre-processing soft-ware,56 specifically developed to handle the different units

and the sensing device constituting the platform, and to per-form the analysis of the collected spectra. Hyperspectralimages have been acquired, every 7 nm, for a total of 121wavelengths. The spectrometer was coupled to a 50-mmlens. The images were acquired scanning the investigated sam-ple line by line. The image width was 320 pixels, while thenumber of frames was 280. Calibration was performed record-ing two images for black and white references. The black image(B) was acquired to remove the effect of dark current of thecamera sensor, turning off the light source and covering thecamera lens with its opaque cup. The white reference image(W) was acquired for a standard white ceramic tile underthe same condition of the raw image. The following equationwas then used to calculate the corrected image (I):

I ¼ I0 − BW − B

100; (1)

where I is the corrected hyperspectral image in a unit of relativereflectance (%); I0 is the original hyperspectral image;B is the black reference image (≈0% reflectance) and W isthe white reference image (≈99.9% reflectance). All thecorrected images were then used as the basis for subsequentanalysis to extract spectral information and for predictionpurposes.

Sample spectra acquisition was carried out on 21 ml ofliquid sample placed in a glass container. Fourteeen hyper-spectral images were acquired from 14 different samplesextracted following the sampling strategy previously out-lined. To improve dataset robustness and models reliability,100 Regions of Interest (ROIs) were then identified in eachhyperspectral collected image (Fig. 5).

Following this approach, collected spectra embed theinformation not only related to the chemical but also tothe physical attributes (e.g., size class characteristics andcontent of the solid fraction, solid/liquid ratio, etc.) of the

SpecIm NIR Spectral Camera™

Energizing (lighting) source

Optic

Sample

Hyperspectral data display

Control panel

Fig. 4 HSI-NIR-based platform utilized to perform GM and digestateproduct characterization.

Fig. 5 False-color-target-image of digestate product in a Petri sampleholder as a result from the hyperspectral acquisition. The white win-dows are representative of the 100 regions of interest (ROIs) utilizedto collect the spectra. Following this approach a full HSI NIR digestatemulti-sampling-continuous-approach, finalized to evaluate a possiblevariation in digestate physical-chemical characteristics, not simpleto evaluate trough a classical-single-sample NIRS based analyticaltechniques, was carried out.

Optical Engineering 111708-5 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 7: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

digestate, with both classes of information influencing diges-tate evolution and biogas production.

3.5 Data Processing

Data analysis was carried out utilizing the PLS Toolbox™ ofEigenvector Research Inc. inside the Matlab™ (MathworksInc.) environment. The partial least squares (PLS) methodwas utilized to build a model able to predict pH, COD,VS, nitrate, and phosphate concentration in the biodegrad-able mixture, starting from the spectra of digestate samples.The choice of PLS regression was due to the fact that it is oneof the most commonly used regression techniques in NIRcalibration.57,58

A dataset of 1400 spectra was used to build a PLS modelthat was carried out using a cross-validation procedure. Forthe given dataset, cross-validation makes a series of experi-ments, each involving the removal of a subset of objects froma dataset (the test set) and the application of this subset datato the model built utilizing the remaining objects in the data-set (the model building set). The contiguous blocks cross-validation (CBCV) procedure was then utilized for thefinal model definition.

An evaluation of the prediction model’s error was alsocarried out. This evaluation helps to better understand theapplicability of the generated model to the dataset not uti-lized during the calibration phase and, as a consequence,its applicability for the digestate to appraise the unknownparameters. The error on the test set was valued by root-mean-square error of cross-validation (RMSECV):

RMSECV ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP

iðyi − yiÞ2nc

;

s(2)

where yi is the prediction value of the sample i in the cali-bration data set; yi is the measured parameter value of samplei in calibration data set, and nc is the number of samples inthe calibration dataset. The calibration model was chosenconsidering the PLS latent variables (LVs) that minimizethe value of the RMSECV, and the higher validation correla-tion coefficient r2. In particular, r2 is the correlation coeffi-cient between the estimated and predicted or calibratedvalues. It is represented by the following expression:

r2 ¼ 1 −P

iðyi − yiÞ2Pi ðyi − yncÞ2

; (3)

where ym is the mean of the value of all samples in the cali-bration set.

4 Experimental TrialsTwo batch digestion trials were carried out:

• Test No. 1: 30.24 kg of WGM and 10 kg of CM weremixed together.

• Test No. 2: 40.86 kg of RGM and 10 kg of CM weremixed together.

In both substrates water was added in order to realize aproduct to digestate with a humidity of about 90% in weight.To realize a full comparison of the results in both tests thesame quantity of inoculum (Table 1) was utilized and the

same mesophilic state (35°C� 2°C) maintained inside thereactor. A continuous monitoring of the process parameters,following both the sampling and the analytical procedurespreviously described was carried out.

5 Results and Discussion

5.1 Digestion Parameters Collection and Analysis

5.1.1 pH

The analysis of pH values, as a function of time, collectedinside the reactor for the two different tests range between 5and 6 (Fig. 6). These values are consistent with those mea-sured by Mallick et al.59 during the anaerobic digestion ofdistillery effluent: the pH value decreased from 7.5 to 5in the first two days and after increased again to 6. The valueswere also confirmed by Fountoulakis et al.,60 who detected apH value between 7.5 and 6.5 during co-digestion of agro-industrial waste, which included winery residues extract.

Tests carried out on WGM presented an initial decrease ofthe pH value. After the fourth day the pH slowly rose. Theearly fall in pH and later recovery is indicative of the acido-genic and methanogenic dominant stages, respectively. Onthe other hand, tests carried out on RGM showed the pHvalue rapidly increased at the beginning, slowly dropped dur-ing the further digestion process and reached stabilization inthe last days of the test. The differing behavior of the pHperformance in the two experiments can be probably attrib-uted to the different extraction process occurring on the twodifferent GMs. As previously outlined, RGM was fermentedbefore its collection, so a small quantity of alcohol was pre-sent. During fermentation, GM acetic acid was produced.Acetic acid was, in fact, the principal substrate convertedin the biogas.61

Tests demonstrated the pH values are strongly linked tothe conversion of dissolved organic matter to fatty acid, andto the further conversion of fatty acids into methane and car-bonic anhydride. The growth of the fatty acid concentrationdrops the pH. The pH value began to gradually rise as theVFA were transformed in methane, indicating depletion ofVFA in the leachate by the methanogenesis.

Fig. 6 pH values collected inside the reactor for the different typesof feedstock during the mesophilic anaerobic digestion process.

Optical Engineering 111708-6 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 8: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

5.1.2 COD trend

The variation of COD concentration in the reactive mixturewas investigated (Fig. 7 and Table 2). The COD presented ahigh concentration value at the beginning of the experiment(COD2 ¼ 58.750 mgO2∕l for WGM and 29.817 mgO2∕lfor RGM, respectively). The values gradually decrease dueto the organic matter intake operated by the micro-organismsfor their growth (Fig. 6). COD values after 14 days were32.500 mgO2∕l for WGM and 15.330 mgO2∕l for RGM,respectively. The degraded substrate was quantified by theremoval yield adopting the following equation:62

XCOD ¼ COD2 − COD14

COD2

· 100. (4)

The overall removal of COD, after the 14-days trial, wasthus equal to 44.68% and 48.59% for WGM and RGMmixture, respectively (Table 2).

The COD values during the reaction are in agreementwith those obtained by Mallick et al.:59 a 45% COD wasachieved after 10 days of anaerobic treatment of distilleryeffluent. Failla and Restuccia63 also confirmed the obtainedresults. In their experiments the COD concentration wasreduced from 85.400 mgO2∕l to 41.500 mgO2∕l after atreatment of 21 days, with a COD removal of 51.4%.COD removal can be correlated to organic matter biodegra-dation and to transformation in biogas. Being the COD anindicator of the quantity of oxygen necessary to oxidize the

Table 2 COD removal performance in WGM and RGM tests.

COD value at2nd day (COD2)

COD value at14th day (COD14)

CODremoval

(mgO2∕l) (mgO2∕l) (%)

White grape marc 58,750 32,500 44.68

Red grape marc 29,817 15,330 48.59

Fig. 7 COD concentrations of the sludge samples collected at differ-ent times inside the reactor during the mesophilic anaerobic digestionprocess.

Fig. 8 Evolution of nitrates concentration in the reactor producedduring the mesophilic anaerobic digestion process.

Table 3 Characteristics of the digestates collected during WGM and RGM experiments.

WGM experiment RGM experiment

Mean Min Max Mean Min Max

Total solid (%WW) 2.12% 1.59% 3.69% 6.32% 1.64% 5.06%

Volatile solid (%TS) 64.10% 56.00% 78.35% 57.56% 51.23% 63.55%

pH 5.41 4.82 6.14 5.45 4.85 6.13

COD (mgO2∕l) 37,607 32,500 58,750 18,579 10,613 29,817

Nitrate (mg∕l) 19.38 9.47 40.14 41.98 9.50 63.23

Fig. 9 Cumulated biogas production as produced in the reactorduring the mesophilic anaerobic digestion process.

Optical Engineering 111708-7 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 9: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

organic matter by chemical reaction,64 a reduction of theCOD concentration in the samples produces a correspondingincrease in biogas production.

5.1.3 Nitrate trend

Nitrate concentration is an important parameter beingdirectly linked to the possibility of reusing the digestate asfertilizer in agriculture. The content of nitrates is thus impor-tant, allowing for the evaluation of the feasibility ofclose-loop-processing: grape production, extraction of thejuice from grapes (for wine production), biogas productionfrom the winery wastes and digestate reuse as fertilizer.Furthermore, a continuous monitoring of nitrate concentra-tion is important in identifying the occurrence of inhibitionphenomena during the digestion process. In fact, in most ofthe studies reported in the literature it was concluded thatnitrate can be considered as a direct inhibitor of methanogensand so in the biogas production.65–67 The evolution of nitrateconcentration in the reactor during the WGM and RGMdigesting process is reported in Fig. 8.

WGM digestion starts with an initial high concentrationof nitrate, which is reduced during the test. In fact, at the

beginning of the experiment the digestate presented a con-centration of about 40.14 mg∕l, concentration that at the endof the digestion process is reduced to about 10.56 mg∕l Inthe RGM digestion, nitrate behavior was similar to WGM. Infact, the digestate presented an initial concentration of nitrateof 62.9 mg∕l that is reduced to 17.85 mg∕l after 14 days.Such trends are justified by the denitrification processmade in the first stage of decomposition (before the metha-nation). The values indicate both the residues from WGMand RGM can be utilized as fertilizers, according to theEuropean Directive 91/676/CEE.

5.1.4 Digestion process performance

Results demonstrated the biogas production rate, as well asthe total produced biogas, is a function of the feedstockorganic content and biodegradability (Table 3): RGM fer-mentation produces a higher quantity of biogas as opposedto the WGM, according to the different level of fermentationof GM to be utilized in the digester (Fig. 9). After 14 days theRGM digestion produced 316.42 l of biogas against 132.60 lproduced by WGM. The evaluation was performed adoptingas an indicative value, the biogas production rate (BPR), that

Fig. 10 Different spectral plots resulting from the different spectra pretreatment: (a) source reflectance spectra of all the analyzed digestate sam-ples, in the MIR field (1000 to 1700 nm). (b) Cut spectra, after peak removal due to the influence of the spectra acquisition set up. (c) Preprocessedspectra applying baseline correction (BC) algorithms. (d) Preprocessed spectra applying BC and absolute value correction (AVC) algorithms.(X axis, investigated wavelength interval (nm); Y axis, reflectance.)

Optical Engineering 111708-8 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 10: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

is the ratio between the biogas produced at the end of thedigestion process and the weight of the dry matter (TS)added at the beginning of each test. BPR values for theRGM and WGM mixtures are equal to 18.79 lBIOGAS∕kgTS and 9.51 lBIOGAS∕kg TS, respectively.

The results achieved in this study are in good agreementwith those obtained by Failla and Restuccia,63 who treatedGM at laboratory small scale, utilizing a digestion architec-ture consisting of 12 digestors of 4l each for 21 days. In thatstudy a production of about 100 lBIOGAS∕kg TS was mea-sured. The difference between the two values can beexplained considering the different substrate ratio used inthe experiment of Failla and Restuccia63 compared to theone of the present paper: inoculum-grape marcs ratio of2∶1 versus an inoculum-grape marcs ratio of 0.12∶1.

5.2 Hyperspectral Approach

5.2.1 Original spectral features and spectralpretreatment

The 1.400 original spectra used to build and validate themodel are reported in Fig. 10. Spectra were quite

homogeneous in shape. Looking at the digestate spectra,as directly acquired and plotted, it is clear that the reflectancevalues at the different wavelengths are correlated to thesampling time sequence in regards to the digestion processevolution. The spectra, in fact, show marked reflectancedifferences (e.g., during the evolution of the anaerobic diges-tion process, spectra intensities decrease), especially in thewavelength region (1300 to 1700) nm. In this region spectralreflectance is influenced by the different water content insidethe samples (e.g., higher water content in the digestate at thebeginning of the process).

Different pre-processing methods were implementedand utilized to improve the robustness of the developedapproach.

The acquired raw spectra were preliminarily “cleaned” toremove the presence of a peak in the wavelength region(1600 to 1700 nm) (Fig. 10). The peak was due to the spec-tral response of the detector in respect to the interaction ofthe energizing source characteristics (Table 4) with the over-all lighting environmental conditions. To reduce the intro-duction of artifacts and to enhance spectral differences abaseline correction (BC) algorithm was preliminarily appliedto all the collected spectra. The absolute value correction(AVC) algorithm was then applied in order to shift the valuesobtained after the baseline correction from negative to posi-tive (Fig. 10). The application of the pre-treatment enhancesthe predictive abilities of the NIR spectra, with referenceto the chemical-physical parameters representative of theanaerobic reaction (Table 5).

Despite the number of the LV increases in the pre-treatment method application, the variation of RMSECVis very significant. The application of pre-treatment methodsto the raw data greatly decreases the RMSECV of thecalibration model of each investigated parameter. Specifi-cally, the RMSEC for COD and nitrate decreased from12861.23 mgO2∕l and 23.2148 mg∕l to 6708.14 mgO2∕land 13.6292 mg∕l. These reductions can be attributed tothe possible correlation existing between the chemical-physical parameter and the NIR spectra. As a result, theaccuracy of the prediction model was increased if a correctpre-treatment of a data set was used, because in this waythe “hide correlations” existing between the spectra andparameter measured were shown.

Table 4 Technical characteristics of the ImSpector™ N17E.

Sensor- Te-cooled InGaAs photodiode array 640 × 512- 14 bit, USB2, LVDS, CameraLink

Spectral range 900 − 1700 nm� 10 nm

Spectralresolution

2.6 nm

Spatialresolution

rms spot radius < 15 μm

Aberrations Insignificant astigmatism, smile or keystone

Effective slitlength

12.8 mm

Numericalaperture

F∕2.0

Stray light <0.5% (halogen lamp, 1400 nm notch filter)

Table 5 Characterization of the 14 digestate samples collected during WGM and RGM experiments by HSI processing based on the applicationof PLS model.

Raw data Processed data

LVs r 2 RMSEC RMSECV LVs r 2 RMSEC RMSECV

COD [mgO2∕l] 15 0.77 7,879.09 8321.14 14 0.90 5,222.89 5,495.16

VS [%TS] 11 0.68 5.85 6.07 13 0.76 5.09 5.29

pH 11 0.67 0.19 0.19 16 0.71 0.17 0.18

Nitrate [mg∕l] 12 0.57 14.19 14.81 13 0.68 12.32 12.83

LVs, latent values; r 2, correlation factor; RMSEC, root-mean-square error of calibration; RMSECV, root-mean-square error of cross-validation.

Optical Engineering 111708-9 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 11: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

5.2.2 HSI-NIR based modeling

The results of the NIR-spectra-collection-based models andthe resulting predictions for the chemical and physical param-eters as collected in the digestate samples are presented inTable 5. Calibration plots of predicted and measured valuesare shown in Fig. 11.

In the COD calibration model, very good results wereobtained (r2 ¼ 0.90 and RMSEC ¼ 5222.89 mgO2∕l),as well as a good validation, since RMSECV was5495.16 mgO2∕l.

Good values were discovered in the pH and nitrate cali-bration models. The r2 values of 0.71 and 0.76 wereobtained, respectively, in pH and VS models. The errorsfor calibration (RMSEC) and validation (RMSECV) werealso acceptable in each parameter (RMSEC ¼ 0.17 andRMSECV ¼ 0.18 for pH model; RMSEC ¼ 17% andRMSECV ¼ 18% for VS model).

The lower value is obtained for nitrate: the model showsan r2 of 0.68. The corresponding RMSEC value is12.32 mg∕l and the RMSECV is 12.83 mg∕l.

The obtained results, even if related to a specific biomasssuch as grape marcs, are quite meaningful especially ifcompared with similar results obtained by other authors,such as Morel et al.68 They built a COD and pH calibrationmodel using spectra obtained by multi-wavelength fluoro-metry obtaining an r2 value of 0.84 and 0.75. Or Zheng-BoYue et al.,69 who built a VS model utilizing a near-infrared-reflectance spectroscopy device obtaining an r2 valueof 0.916.

6 ConclusionsThe tests carried out on the white and red grape marc, WGMand RGM, allowed for achieving several research goals, eachlinked to the potentialities of this “particular” agro-industrialbyproduct to be profitably utilized in biogas production andthe possibility of utilizing an hyperspectral approach in theNIR (1000 to 1700-nm) range to perform a preliminary mod-eling of the process and to allow developing control strate-gies for a full monitoring of the process itself.

With reference to biogas production, the followingassumption can be made:

• RGM digestion is “easier” if compared to WGM.In fact, BPR value related to RGM processingis bigger (18.79 lBIOGAS∕kg TS) than WGM(9.51 lBIOGAS∕kg TS);

• the larger percentage of COD removal in an RGMsubstrate, with respect to the WGM, is explained bythe larger quantity of biogas resulting from RGMdegradation in respect to the WGM;

• a winery waste pre-treatment (alcohol and other inhi-bitor factors dissolved in the original organic matterremoval) could improve the process performance.

With reference to the utilization of the hyperspectralapproach for process modeling/control:

• results showed a HSI-NIR-based prediction can consti-tute a feasible technique to rapidly and accurately

Fig. 11 Calibration plots showing the correlation existing between predicted and measured values. (a) COD, (b) VS, (c) pH, and (d) nitrate.

Optical Engineering 111708-10 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 12: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

quantify several anaerobic digestion-process-related-parameters as COD, pH, VS, and nitrate;

• HSI-NIR could be profitably utilized to perform a fullreal-time monitoring of an anaerobic digestion process.Following this approach, the possibility to build HSI-NIR-based prediction models can also reduce the costslinked to classical off-line-batch chemical and physicalanalyses;

• the proposed HSI strategies could thus allow for carry-ing out a real-time identification of possible reductionsin biogas production due to some inhibitory events inthe anaerobic reaction and to set up innovative controlstrategies that allow for performing full continuousmonitoring and control of the process.

Further studies will be carried in order to investigate awider utilization of GMs for biogas production with refer-ence to different grapes originating from different GMs andwine wastewater, according to different vinification processand GM storage; the possibility to develop a co-digestionprocess based on different GM-based recipes; to develop aninnovative HSI-NIR-based architecture (e.g., sensing probesand algorithms) that is able to perform a continuous collec-tion and processing of the information related to digestatestatus and optimal biogas production.

AcknowledgmentsThe authors wish to thank Eng. Aldo Gargiulo and Dr. LauraD’Aniello of Sapienza University of Rome for their technicalsupport during the development of the work.

References

1. R. Moletta, D. Verrier, and G. Albagnac, “Dynamic modeling ofanaerobic-digestion,” Water Res. 20(4), 427–434 (1986).

2. A. Wheatley, Anaerobic Digestion. A Waste Treatment Technology,Elsevier Applied Science, London (1990).

3. A. J. Ward et al., “Optimisation of the anaerobic digestion of agricul-tural resources,” Bioresour. Technol. 99(17), 7928–7940 (2008).

4. C. Dumas et al., “Combined thermophilic aerobic process and conven-tional anaerobic digestion: effect on sludge biodegradation and methaneproduction,” Bioresour. Technol. 101(8), 2629–2636 (2010).

5. J. Mata-Alvarez, “Biomethanization of the organic fraction of municipalsolid wastes,” 1st ed., IWA Publishing, London (2002).

6. B. Lund et al., “Inactivation of virus during anaerobic digestion of man-ure in laboratory scale biogas reactors,” Antonie Van Leeuwenhoek69(1), 25–31 (1996).

7. L. Sahlstrom, “A review of survival of pathogenic bacteria in organicwaste used in biogas plants,” Bioresour. Technol. 87(2), 161–166(2003).

8. E. Smet, H. Van Langenhove, and I. De Bo, “The emission of volatilecompounds during the aerobic and the combined anaerobic/aerobiccomposting of biowaste,” Atmos. Environ. 33(8), 1295–1303 (1999).

9. S. T. Wagland et al., “Test methods to aid in the evaluation of thediversion of biodegradable municipal waste (BMW) from landfill,”Waste Manag. 29(3), 1218–1226 (2009).

10. S. Tafdrup, “Viable energy production and waste recycling fromanaerobic digestion of manure and other biomass materials,” BiomassBioenergy 9(1–5), 303–314 (1995).

11. M. Pöschl, S. Ward, and P. Owende, “Evaluation of energy efficiencyof various biogas production and utilization pathways,” Appl. Energy87(11), 3305–3321 (2010).

12. R. Moletta, Management of environmental issues in food industry(Gestion des Problèmes Environnementaux Dans les Industries Agro-alimentaires), 2nd ed., Tec & Doc, Paris, France (2006).

13. B. P. Kelleher et al., “Advances in poultry litter disposal technology—areview,” Bioresour. Technol. 83(1), 27–36 (2002).

14. ISRPA. Rapporto Rifiuti Urbani 2009 “Italian Institute for the environ-ment protection and research,” Annual report (2009).

15. K. A. Bjorndal and J. E. Moore, “Prediction of fermentability ofbiomass feedstocks from chemical characteristics,” in Biomass EnergyDevelopment, W. H. Smith, Ed., pp. 447–454, Plenum Press, New York(1985).

16. J. A. Chandler et al., “Predicting methane fermentation biodegradabil-ity,” Biotechnol. Bioeng. Symp. 10, 93–107 (1980).

17. V. N. Gunaseelan, “Regression models of ultimate methane yields offruits and vegetable solid wastes, sorghum and napiergrass on chemicalcomposition,” Bioresour. Technol. 98(6), 1270–1277 (2007).

18. Y. W. Han, J. S Lee, and A. W. Anderson, “Chemical composition anddigestibility of ryegrass straw,” J. Agric. Food Chem. 23(5), 928–931(1975).

19. C. Habig, “Influences of substrate composition on biogas yields ofmethanogenic digesters,” Biomass 8(4), 245–253 (1985).

20. C. Allison and G. T. Macfarlane, “Effect of nitrate on methaneproduction and fermentation by slurries of human faecal bacteria,”J. Gen. Microbiol. 134(6), 1397–1405 (1988).

21. D. Deublein and A. Steinhauser, “Biogas from waste and renewableresources: an introduction,” Wiley-VCH, USA (2008).

22. H. V. Hendriksen and B. K. Ahring, “Integrated removal of nitrate andcarbon in an upflow anaerobic sludge blanket (UASB) reactor: operat-ing performance,” Water Res. 30(6), 1451–1458 (1996).

23. J. C. M. Scholten and A. J. M. Stams, “The effect of sulfate and nitrateon methane formation in a freshwater sediment,” Antonie van Leeuwen-hoek 68(4), 309–315 (1995).

24. D. Cozzolino, “Chemometrics and visible-near infrared spectroscopicmonitoring of red wine fermentation in a pilot scale,” Biotechnol.Bioeng. 95(6), 1101–1107 (2006).

25. S. Macho and M. S. Larrechi, “Near-infrared spectroscopy andmultivariate calibration for the quantitative determination of certainproperties in the petrochemical industry,” Trends Anal. Chem. 21(12),799–806 (2002).

26. B. G. Osvorne, “Near infrared spectroscopy in food analysis”, in Ency-clopedia of Analytical Chemistry, Wiley (1986).

27. S. Vaidyanathan, “Assessment of near-infrared spectral informationfor rapid monitoring of bioprocess quality,” Biotechnol. Bioeng. 74(5),376–388 (2001).

28. J. B. Holm-Nielsen, C. K. Dahl, and K. H. Esbensen, “Representativesampling for process analytical characterization of heterogeneousbioslurry systems—a reference study of sampling issues in PAT,”Chemometr. Intell. Lab. Syst. 83(2), 114–126 (2006).

29. A. Schievano et al., “Predicting anaerobic biogasification potentialof ingestates and digestates of full-scale biogas plant using chemicaland biological parameters,” Bioresour. Technol. 99(17), 8112–8117(2008).

30. M. Lesteur et al., “First step towards a fast analytical method for thedetermination of biochemical methane potential of solid wastes bynear infrared spectroscopy,” Bioresour. Technol. 102(3), 2280–2288(2011).

31. A. A. Gowen et al., “Hyperspectral imaging—an emerging processanalytical tool for food quality and safety control,” Trends Food Sci.Technol. 18(12), 590–598 (2007).

32. D. W. Sun, Ed., Hyperspectral Imaging for Food Quality Analysis andControl, Academic Press/Elsevier, Ed., San Diego (2010)

33. A. Del Fiore et al., “Early detection of toxigenic Fungi on maize byhyperspectral imaging analysis,” Int. J. Food Microbiol. 144(1),64–71 (2010).

34. A. A. Gowen et al., “Recent applications of chemical imaging to phar-maceutical process monitoring and quality control,” Eur. J. Pharm.Biopharm. 69(1), 10–22 (2008).

35. W. Fortunato de Carvalho Rocha et al., “Quantitative analysis of pirox-icam polymorphs pharmaceutical mixtures by hyperspectral imagingand chemometrics,” Chemometr. Intell. Lab. Syst. 106(2), 198–204(2011)..

36. R. Jolivot, P. Vabres, and F. Marzani, “Reconstruction of hyperspectralcutaneous data from an artificial neural network-based multispectralimaging system,” Comput. Med. Imag. Graph. 35(2), 85–88 (2011).

37. Z. Liu et al., “Classification of hyperspectral medical tongue images fortongue diagnosis,” Comput. Med. Imag. Graph. 31(8), 672–678 (2007).

38. M. Kubik, “Hyperspectral imaging: a new technique for the non-invasive study of artworks,” Chapter 5 in Physical Techniques in theStudy of Art, Archaeology and Cultural Heritage, D. Creagh andD. Bradley, Ed., pp. 199–259, Elsevier (2007).

39. R. Gosselin, D. D Rodrigue, and C. Duchesne, “A hyperspectralimaging sensor for on-line quality control of extruded polymer compo-site products,” Comput. Chem. Eng. 35(2), 296–306 (2011).

40. P. Tatzer, M. Wolf, and T. Panner, “Industrial application for inlinematerial sorting using hyperspectral imaging in the NIR range,”Real-Time Imag. 11(2), 99–107 (2005).

41. G. Bonifazi and S. Serranti, “Imaging spectroscopy based strategies forceramic glass contaminants removal in glass recycling,” Waste Manag.26(6), 627–639 (2006).

42. G. Bonifazi and S. Serranti, “Hyperspectral imaging based techniquesin fluff characterization,” Proc. SPIE Optic East 2006b; 6377,Advanced Environmental, Chemical and Biological Sensing Tech-nologies IV, October 3–4, Seaport World Trade Center, Boston,Massachusetts.

43. G. Bonifazi and S. Serranti, “Hyperspectral imaging based proceduresapplied to bottom ash characterization,” Proc. SPIE Optic East 2007;6755, Advanced Environmental, Chemical and Biological Sensing

Optical Engineering 111708-11 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms

Page 13: Hyperspectral imaging techniques applied to the monitoring of wine waste anaerobic digestion process

Technologies V, September 10–11, Seaport World Trade Center, Boston,Massachusetts.

44. Serranti S et al., “Composting products quality assessment and moni-toring by hyperspectral imaging based logics,” in Proc. Waste-to-Resources, III International Symp. MBT&MRF, May 12–14, Hanover,Germany, pp. 584–597 (2009).

45. R. Leitner, H. Mairer, and A. Kercek, “Real-time classification of poly-mers with NIR spectral imaging and blob analysis,” Real-time Imag.9(4), 245–251 (2003).

46. S. Serranti, A. Gargiulo, and G. Bonifazi, “Characterization of post-consumer polyolefin wastes by hyperspectral imaging for quality con-trol in recycling processes,” Waste Manag. 31(11), 2217–2227 (2011).

47. S. Serranti, A. Gargiulo, and G. Bonifazi, “Classification of polyolefinsfrom building and construction waste using NIR hyperspectral imagingsystem,” Resour. Conservat. Recycl. 61, 52–58 (2012).

48. OIV—International organizaton of Vine and Wine, “Structure of theworld vitivinicultural industry in 2007,” Report (2007).

49. I. S. Arvanitoyannis, D. Lads, and A. Mavromatis, “Wine waste treat-ment methodology,” Int. J. Food Sci. Technol. 41(10), 1117–1151(2006).

50. S. Cortés et al., “The storage of grape marc: limiting factor in the qualityof the distillate,” Food Control. 21(11), 1545–1549 (2010).

51. A. M. Gonzalez-Paramas et al., “Flavanol content and antioxidantactivity in winery by-products,” J. Agric. Food Chem. 52(2), 234–238(2004).

52. F. Tambone et al., “Assessing amendment properties of digestate bystudying the organic matter composition and degree of biological sta-bility during the anaerobic digestion of the organic fraction of MSW,”Bioresour. Technol. 100(12), 3140–3142 (2009).

53. S. Ghosh and E. R. Vieitez, “Biogasification of solid waste by two phaseanaerobic fermentation,” Biomass Energy 16(5), 299–309 (1999).

54. IRSA-CNR. “Metodi analitici per le acque,” Manuali e Linee Guida29/2003 (2003),www.irsa.cnr.it/Docs/Capitoli/1000.pdf

55. J. A. Álvarez, L. Otero, and J. M. Lema, “Amethodology for optimisingfeed composition for anaerobic co-digestion of agro-industrial wastes,”Biotechnol. Res. 101(4), 1153–1158 (2010).

56. SSOM, Spectral scanner operative manual (Version 2.0). DVOptics Srl,Italy (2008).

57. E. Dabakk et al., “Inferring lake water chemistry from filtered sestonusing NIR spectrometry,” Water Res. 34(5), 1666–1672 (2000).

58. P. Geladi and E. Daebakk, “An overview of chemometrics applicationsin NIR spectroscopy,” J. Near Infrared Spectrosc. 3(1), 119–132(1995).

59. P. Mallick, J. C. Akunna, and G. M. Walker, “Anaerobic digestion ofdistillery spent wash: influence of enzymatic pre-treatment of intactyeast cells,” Bioresour. Technol. 101(6), 1681–1685 (2010).

60. M. S. Fountoulakis et al., “Potential for methane production from typi-cal Mediterranean agro-industrial by products,” Biomass Energy 32(2),155–161 (2008).

61. H. Boualagui et al., “Bioreactor performance in anaerobic digestionof fruit and vegetable wastes,” Process Biochem. 40(3–4), 989–995(2005).

62. F. Benitez et al., “Aerobic and anaerobic purification of wine distillerywastewater in batch reactors,” Chem. Eng. Technol. 22(2), 165–172(1999).

63. S. Failla and A. Restuccia, “Impiego delle vinacce per scopi energetici:Prime valutazioni con un impianto da laboratorio,” IX ConvegnoNazionale dell’Associazione Italiana di Ingegneria Agraria, pp. 10–13(2009).

64. G. Tchobanoglous, F. L. Burton, and H.D. Stensel, Wastewater Engi-neering, Treatment, Disposal And Reuse, 4th ed. McGraw-Hill Inc,New York (2006).

65. C. Allison and G. T. Macfarlane, “Effect of nitrate on methane produc-tion and fermentation by slurries of human faecal bacteria,” J. Gen.Microbiol. 134(6), 1397–1405 (1988).

66. J. C. M. Scholten and A. J. M. Stams, “The effect of sulfate and nitrateon methane formation in a freshwater sediment,” Antonie Van Leeuwen-hoek 68(4), 309–315 (1995).

67. H. V. Hendriksen and B. K. Ahring, “Integrated removal of nitrate andcarbon in an up-flow anaerobic sludge blanket (UASB) reactor: operat-ing performance,” Water Res. 30(6), 1451–1458 (1996).

68. E. Morel et al., “Application of multi-wavelength fluorometry foron-line monitoring of an anaerobic digestion process,” Water Res.38(14–15), 3287–3296 (2004).

69. Z.-B. Yue et al., “Determination of main components and anaerobicrumen digestibility of aquatic plants in vitro using near-infrared-reflectance spectroscopy,” Water Res. 44(7), 2229–2234 (2010).

Silvia Serranti is assistant professor at theDepartment of Chemical Engineering, Mate-rials & Environment (DICMA), Faculty of Civiland Industrial Engineering, University ofRome “La Sapienza.” She is a PhD geologistand has been working for 15 years in the RawMaterials Unit of DICMA. Her research activ-ity is related to the field of primary and sec-ondary raw materials characterization andvalorization, and is documented by morethan 100 scientific papers published in inter-

national journals and in conference proceedings and by participatingin 10 research projects financed by the European Union. Her maininterest topics include the study of particulate solids, with referenceto their dimensional, morphological, and morphometrical attributesand physical-chemical characteristics; identification of objects usingpattern recognition techniques based on image analysis; innovativeprocedures for characterization, recognition, classification, and sort-ing of primary and secondary raw materials by hyperspectral imagingtechniques; the development of sensing methodologies for on-linesorting of different materials and/or quality control of industrial pro-cesses and final products; airborne dust measurement and character-ization; froth flotation monitoring in industrial plants by digital imageprocessing; satellite-image-based strategies to evaluate the impactof dismissed mine sites.

Andrea Fabbri is a PhD student in theDepartment of Chemical Engineering, Mate-rials & Environment (DICMA), Faculty ofCivil and Industrial Engineering, Universityof Rome “La Sapienza.” His graduate thesisin 2009 concerned the technical, energeticand economical application of anaerobicdigestion to municipal organic solid wasteproduced in Latina Country (Italy). He wona prize for the best thesis in business innova-tion in 2010. His research activity is related

to the energetic and materials valorization of organic waste fromthe agricultural sector. His main topics of interest include the studyof organic waste in regards to their industrial/agricultural output;biomethane potential estimation of organic waste investigatedusing standard methods; monitoring of the behaviors of chemical andphysical parameters during the anaerobic digestion process; develop-ment of innovative procedures for chemical and physical characteri-zation of organic materials by hyperspectral imaging techniques andof digestate collected from the anaerobic reactor during the digestionprocess.

Giuseppe Bonifazi is professor of rawmaterials beneficiation at the Department ofChemical Engineering Materials and Envi-ronment at La Sapienza - University of Rome.He has more than 25 years experience on thecharacterization of particles and particulatesolids materials by image processing. Hismain scientific and technical fields of investi-gation are related to the study of software andhardware integrated architectures for thesynthesis, classification, and recognition of

numeric signals; the development and set-up of procedures for theidentification of objects and materials using pattern recognition tech-niques based on classical and hyperspectral imaging techniques;and analysis and application of methodologies to study and modelindustrial processes with reference to particulate solids.

Optical Engineering 111708-12 November 2012/Vol. 51(11)

Serranti, Fabbri, and Bonifazi: Hyperspectral imaging techniques : : :

Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 09/30/2013 Terms of Use: http://spiedl.org/terms