This PDF file contains the front matter associated with SPIE Proceedings Volume 10665, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and Conference Committee listing.

In modern agriculture drought is a major cause of low yields worldwide. Therefore, imaging systems that enable to study the interactions between plant and soil are a way towards better understanding of crop water supply. The main objective of root measurement is to obtain new knowledge on key functioning of roots. An important aspect is that the roots should be measured in their natural soil environment to obtain information not only on root architecture but also on root functioning. Within this research project a VIS imaging setup and a NIR hyperspectral imaging system for the acquisition of hyper-spectral NIR image data of rhizoboxes were developed. The combination of imaging using the VIS wavelength range (380nm to 780nm) and imaging spectroscopy using the NIR wavelength range (900nm to 1700nm) provides several advantages. The VIS imaging setup is used to provide quick overview images at different development states of plant roots. The hyperspectral NIR system provides increased image contrast which allows for a more reliable segmentation of the roots from the soil and additional information, e.g. basic root-architecture, to be extracted. Moreover, it is possible to visualize the water distribution in the soil in close proximity to the roots. In this paper the hardware setup of the NIR imaging spectroscopy system, the data analysis, the soil water distribution measurements and the root segmentation are presented.

Applications of machine vision techniques are prevalent for quality inspection of foods. For safety inspection of fruits such as apples to detect biological contaminants, a method to capture and reconstruct a whole-surface of apple is needed. In this paper, we present a reconstruction method for whole-surface imaging of apples with the use of a line-scan hyperspectral imaging technique. In addition, the development of online whole-surface inspection technology for round-fruits is presented.

Continuous gradient temperature Raman spectroscopy (GTRS) is a simple, rapid technique for determining the unique structures of fatty acids, triacylglycerols and phospholipids. Improved lipid spectra and vibrational assignments have many applications in food safety and quality. Herein we analyze the residual lipid components in porcine and poultry meat and bone meal (MBM) collected directly from rendering operations. We are developing a rapid throughput GTRS method that requires no special extraction methods or toxic/expensive solvents, and can be adapted to field use. Crude ethanol, methanol and water extracts of pork, poultry samples and 20:80 pork:poultry MBM were investigated from -100 to 80°C. GTRS provides 20 Mb three-dimensional data arrays with 0.2°C increments and graphical first and second derivatives. Comparison of second derivative data showed good reproducibility among samples, with some vibrational modes distinct for either pork or chicken. The 20:80 pork:poultry data showed for the first time that lipid prepared from mixed MBM can be positively identified. Oil from pollack waste was also examined. GTRS and other methods can easily identify pure fishmeal or fishmeal mixed into terrestrial MBM; fishmeal is distinguished by its long-chain polyunsaturated fatty acid content. The analytical challenge is to determine economic or accidental adulteration whereby porcine, bovine or ovine MBM are mixed with each other, or mixed into poultry or fish meals.

Low temperature environment affects the growth and yield of watermelon negatively. The conventional visual inspection with human eyes has a limitation for accurate phenotyping of the stress symptom. Spectral imaging technique has been used as a useful phenotyping tool for visualizing physiological responses of plants. In this study, responses of chilling stresses of watermelon leaves were investigated using Vis/NIR hyperspectral imaging (HSI) technique. Sensitive and resistant to chilling tolerance of watermelon plants were exposed to low temperature conditions. HSI of the treated leaves were collected and analyzed with multivariate analysis. The result shows that HSI technique could distinguish between susceptible and resistant plants.

The objective of this study was to predict the moisture content, soluble solids content, and titratable acidity content in bell peppers during storage, based on hyperspectral imaging (HSI) in the 1000–1500 nm wavelength range. The mean spectra of 148 mature bell peppers were extracted from the hyperspectral images, and multivariate calibration models were built using partial least squares regression to predict MC, SSC, and TA content in bell peppers with different preprocessing techniques. The selected optimum wavelengths were used to create distribution maps for MC, SSC, and TA content of bell peppers. The results revealed that HSI coupled with multivariate analysis can be used successfully to predict the MC, SSC, and TA content in bell peppers.

We present a multimode hyperspectral imaging (HIS) system operating in fluorescence and reflectance modes for food quality and safety applications. The system uses spectral band sequential imaging on the detection side. To ensure constant, high S/N levels, the image acquisition time is optimized for each spectral band. The illumination module uses two independent light sources for fluorescence and reflectance measurements, based on three computer-controlled LED illumination rings. UVA (371 nm, FWHM 16nm, power = 91.7mW/cm²) and blue/violet (418 nm, FWHM = 21nm, power = 38.9mW/cm²) LEDs provide fluorescence excitation. White LEDs (power = 35.8mW/cm²) are used for reflectance. The spectral imaging system incorporated within the detection pathway is able to transition between wavelengths within microseconds over the full bandwidth of the device (450 nm - 800 nm). The system is configured as a tabletop platform with both the illumination and detection located above the food sample. Illumination uniformity is ~90%, spatial resolution is 89μm, and spectral resolution is 8nm. The system was tested for food safety applications by imaging of pet food spiked with Salmonella enterica, where the ability to identify the bacteria in these samples was compared to existing detection methodologies. As an example of food authentication applications, images captured at 75 wavelengths over the range 450 nm to 810 nm with a 5 nm interval were collected from wild salmon and farmed salmon purchased locally. Wild salmon and farmed salmon were found to have distinctly different reflectance features from 515 nm to 650 nm.

Mandarin orange is a popularly consumed fruit in Asian countries. Over 99% of cultivation area in Korea for mandarin oranges is concentrated in Jeju Island. Despite of this high concentration, detecting infection and estimating fruit yields has been done manually, resulting in loss of money and time. In this study, hyperspectral fluorescence imaging technique was explored to distinguish green mandarin oranges from leaves to estimate fruit density. In addition, early stage detection for disease infection of leaves and fruits were investigated. The fluorescence spectral images showed reliable performance for distinguishing green mandarin oranges from leaves, and detecting disease infection on both leaves and fruits. The result demonstrated that hyperspectral fluorescence imaging might be used for rapid and non-destructive detection of disease infection and yield estimation of mandarin orange in the field.

Spice powders are used as food additives for flavor and color. Economically motivated adulteration of spice powders by color dyes is hazardous to human health. This study explored the potential of a 1064 nm Raman chemical imaging system for identification of azo color contamination in spice powders. Metanil yellow and Sudan-I, both azo compounds, were mixed separately with store-bought turmeric and curry powder at the concentration ranging from 1 % to 10 % (w/w). Each mixture sample was packed in a shallow nickel-plated sample container (25 mm x 25 mm x 1 mm). One Raman chemical image of each sample was acquired across the 25 mm x 25 mm surface area using a 0.25 mm step size. A threshold value was applied to the spectral images of metanil yellow mixtures (at 1147 cm^-1) and Sudan-I mixtures (at 1593 cm^-1) to obtain binary detection images by converting adulterant pixels into white pixels and spice powder pixels into the black (background) pixels. The detected number of pixels of each contaminant is linearly correlated with sample’s concentration (R² = 0.99). This study demonstrates the 1064 nm Raman chemical imaging system as a potential tool for food safety and quality evaluation.

Although many studies have been conducted to detect melamine in milk powder using near-infrared hyperspectral imaging system, the reproducibility due to moisture content in powder sample and detection limit have not been addressed appropriately. The objective of this study is to develop, based on shortwave infrared (SWIR) hyperspectral imaging, optimal model which is less sensitive to change of moisture content in sample powder. The hyperspectral imaging system consists of a MCT-based camera capable of measuring wavelengths from 1000nm to 2500nm. A halogen-based light source module was used to illuminated samples. The results showed a mixture concentration as low as 50 ppm of melamine in milk could be detected. The detection accuracy using the wavelength region from 1700nm to 2500nm was higher than that of using the wavelength from 1000nm to 1700nm. The MCT-based SWIR hyperspectral imaging system has a good potential for the detection and quantification of adulterants in powder sample.

Rice blast (Magnaporthe oryzae) is one of the most devastating rice diseases affecting yield and grain quality. Applications of Fourier-Transform Infrared (FTIR) spectroscopy to detect plant stress responses to pathogens have been widely investigated; however, assessing the difference in basal resistance (PAMP-triggered immunity; compatible reaction) vs. effector-triggered immunity (incompatible reaction) has remained largely elusive. Here, we inoculated 2-week old rice seedlings (varieties Dee Gee Woo Gen and Lemont) with rice blast isolates IC17 (compatible isolate) and IB54 (incompatible isolate). Leaf tissues were collected at 6, 24, 48, 72 hours, and 7 days after inoculation, and then the constituents were extracted with dimethyl sulfoxide for FTIR analysis. Overall, distinctive profiles of compatible and incompatible reactions compared to the controls were observed. The preliminary result suggests a potential application of FTIR spectroscopy as a means for discriminating the basal resistance and effector-triggered immunity.

This paper investigates the accuracy of surface-scanning measurement of a wireless magnetoelastic (ME) biosensor for direct pathogen detection on solid surfaces. The model experiments were conducted on the surface of a at polyethylene (PE) plate. An ME biosensor (1 mm x 0.2 mm x 30 µm) was placed on the PE surface, and a surface-scanning detector was aligned to the sensor for wireless resonant frequency measurement. The position of the detector was accurately controlled by using a motorized three-axis translation system (i.e., controlled X, Y, and Z positions). The results showed that the resonant frequency variations of the sensor were -125 to +150 Hz for X and Y detector displacements of ± 600 µm and Z displacements of +100 to +500 µm. These resonant frequency variations were small compared to the sensor's initial resonant frequency (˂ 0.007% of 2.2 MHz initial resonant frequency) measured at the detector home position, indicating high accuracy of the measurement. In addition, the signal amplitude was, as anticipated, found to decrease exponentially with increasing detection distance (i.e., Z distance). Finally, additional experiments were conducted on the surface of cucumbers. Similar results were obtained.

In this paper, a novel device named as phage filter is designed and presented to capture and identify a small number of Salmonella Typhimurium cells from large volumes of water. This phage filter is constructed from a filter chamber, filter frames on a spindle, strip-shape magnetoelastic filter elements, and a spinning speed control unit. The filter elements are made from Metglas 2826MB and coated with a specifically designed phage that only binds with Salmonella Typhimurium. These phage-coated filter elements can be held and arranged on the filter frames by magnetic force produced from couples of permanent magnets in the frame. Layers of filter frames are fixed on the spindle. The spindle with filter frames and filter elements can spin in the filter chamber and the spinning speed can be continuously adjusted. When the filter works, the tested water passes through the filter frame, and Salmonella Typhimurium cells striking on the filter elements can be bound by the phage on the element surfaces and removed from the tested water.

To perform rapid sensing of pathogens on the surface of food or food preparing plates, ME wireless biosensing system was combined with surface swab sampling techniques in this research. The ME biosensors which consist of ME resonators E2 phage was generally used for Salmonella typhimurium direct detections on the surfaces. E2 phage used in this research was designed for Salmonella typhimurium specific binding. Instead of measuring one spot at a time, the desired area or the whole area of a target surface can be swabbed for the inexpensive, rapid and easy-to-use pathogen collections. In this study, we first investigated the efficiency of capture and release of a model pathogen, Salmonella Typhimurium, by swab sampling on wet or dry surfaces. Plate counting was used to identify the recovery rates. The efficiency of capture and release was calculated and compared between various kinds of swabs which were composed of different tip materials, including cotton, rayon, and nylon-flocked ones.

The current method for the extraction of olive oil consists on the use of a decanter to split it by centrifugation. During this process, different olive oil samples are analyzed in a chemical laboratory in order to determine moisture levels in the oil, which is a decisive factor in olive oil quality. However, these analyses are usually both costly and slow. The developed prototype is the foundation of an instrument for real-time monitoring of moisture in olive oil. Using the olive oil as the dielectric of a parallel-plate capacitor, a model to relate the moisture in olive oil and capacitance has been created. One of the challenges for this application is the moisture range, which is usually between 1 and 2%, thus requiring the detection of pF-order variations in capacitance. This capacitance also depends on plate size and the distance between plates. The presented prototype, which is based on a PSoC (Programmable System-on-Chip), includes a reconfigurable digital and analog subsystem, which makes the determination of moisture independent of the capacitor. Finally, the measure is also sent to a smartphone via Bluetooth.

Economically motivated adulteration and fraud to food powders are emerging food safety risks that threaten the health of the general public. In this study, targeted and non-targeted methods were developed to detect adulterants based on macro-scale Raman chemical imaging technique. Detection of potassium bromate (PB) (a flour improver banned in many countries) mixed in wheat flour was used as a case study to demonstrate the developed methods. A line-scan Raman imaging system with a 785 nm line laser was used to acquire hyperspectral image from the flour-PB mixture. Raman data analysis algorithms were developed to fulfill targeted and non-targeted contaminant detection. The targeted detection was performed using a single-band Raman image method. An image classification algorithm was developed based on single-band image at a Raman peak uniquely selected for the PB. On the other hand, a mixture analysis and spectral matching method was used for the non-targeted detection. The adulterant was identified by comparing resolved spectrum with reference spectra stored in a pre-established Raman library of the flour adulterants. For both methods, chemical images were created to show the PB particles mixed in the flour powder.

The advancement of the broiler industry in meat processing efficiency and production yield is remarkable. However, the industry has also experienced an emerging meat quality defect, called wooden breast syndrome. The symptoms of wooden breast syndrome include hardened muscle, pale color, ridge-like bulging, connective tissue accumulation, and/or rubbery texture. This study is concerned with the latest research progress within USDA-ARS to develop real-time machine vision system for rapid online detection of wooden breast fillets in the broiler industry. Because the current industry method of wooden breast detection is through tactile evaluation and product handling by humans, a rapid and non-invasive sensing technique to detect meat products affected by wooden breast syndrome is invaluable to both the industry and the scientific community. The developed machine vision system was designed to detect breast fillets moving on a conveyor belt system and differentiate between normal and wooden breast fillets. The imaging system captures and analyzes the physical properties that are correlated with severity of wooden breast condition. The machine vision system consists of a digital CMOS camera, a lighting system, a computer, and software. Shape descriptors characterizing differences between contours of normal and affected breast fillets were developed. Preliminary results obtained with 45 fillets (15 normal, 15 moderate wooden breast, and 15 severe wooden breast) indicated 98 % overall accuracy with a 6.7% false positive rate for normal fillets. A discussion for its commercialization is ongoing with an industry partner.

Despite the complexity of the factors that lead to loss of seed viability, conventional methods like germination tests, tetrazolium tests are commonly employed to determine it. However, these methods have downsides like being destructive, time consuming and non-representative. Therefore, there is a need to develop a fast, non-destructive and real-time measurement and sorting system of seeds based on viability for industrial purpose. In this study, we seek to utilize HSI and multivariate data analysis techniques to classify viable seeds from non-viable ones and later use it basis to develop an online real-time detection system for sorting these seeds based on viability. For this cause, Data from melon and watermelon seeds were collected using a SWIR HSI system. The performance of the classification models achieved both during calibration and real-time tests were quite impressive and a proof that HSI can be effectively applied to an industrial real-time sorting system.

Aflatoxins are fungal toxins produced by Aspergillus flavus. Food and feed crops get contaminated with carcinogenic aflatoxins, which often results in economic losses as well as serious health issues. Grain elevators need to unload, on average, one 50,000-pound truckload every two minutes. Current chemical and optical methods for aflatoxin detection cannot meet the screening requirements. Therefore, a high speed batch screening system with reliable accuracy is necessary. The contaminated corn kernels were prepared in our laboratory by artificial inoculation of corn ears. One hundred 200g samples were selected for analysis. To develop a high speed multispectral screening system, two high performance cameras in conjunction with dual UV excitation sources and novel image processing software were utilized to collect fluorescence images of each sample. Each camera simultaneously captures a single band fluorescence image (436 nm and 532 nm) from corn samples, and the detection software processes the images to automatically detect contaminated kernels by using a normalized difference fluorescence index. Each sample was imaged/screened four times, and screened samples were chemically analyzed for aflatoxin content. All samples were shuffled between imaging repetitions to increase the likelihood of screening both sides of every kernel. Processing time for each screening was about 0.7s, and an optimal result of 98.65% was achieved for sensitivity and 96.6% for specificity.

Aflatoxin contamination can occur in a wide variety of agricultural products pre- and post-harvest, posing potential severe health hazards to human and livestock. However, current methods for detecting aflatoxins are generally based on wet chemical analyses, which are time-consuming, destructive to test samples and require skilled personnel to perform, making them impossible for large-scale non-destructive screening and on-site detection. In this study, we utilized the visible/near-infrared (Vis/NIR) spectroscopy over the spectral range of 400-2500 nm to detect contamination of shelled commercial peanut kernels with the predominant aflatoxin B₁ (AFB₁). Our results indicated the usefulness of Vis/NIR spectroscopy combined with the chemometric techniques of partial least squares discriminant analysis (PLS-DA) and least squares support vector machine (LS-SVM) in identifying the AFB₁ contamination of peanut kernels. Both PLS-DA and LS-SVM methods provided satisfactory classification results using the full spectral information over the ranges of 410-1070 (I), 1120-2470 nm (II) and I+II. Based on the classification threshold of 20 ppb, the best PLS-DA prediction results using the full spectra yielded the average accuracy of 87.9% and overall accuracy of 88.6%. With 100 ppb as the classification threshold, the best PLS-DA model using the full spectra achieved the average accuracy of 94.0% and overall accuracy of 91.4%. Using the full spectra, the best average accuracies recorded by LS-SVM were 90.9% and 98.0%, with the classification thresholds of 20 and 100 ppb, respectively. Correspondingly, the best overall accuracies by LS-SVM were 90.0% and 97.1%. In addition, the simplified models of CARS-PLS-DA and CARS-LS-SVM also demonstrated good prediction capability in identifying the AFB1 contamination from peanut surface. Based on both classification thresholds of 20 and 100 ppb, the best CARS-PLS-DA and CARS-LS-SVM prediction results were ≥ 90.0% in both average accuracy and overall accuracy. Most importantly, the computation complexity and the employed data dimensionality were significantly reduced by using the simplified models.

This research is framed within FoodIntegrity, EU sponsored project(7th FP). The main goal of the research to be done is to provide industrials, producers and consumers with a methodology based in low-cost, portable and miniature NIRS sensors and information and communication technologies for process control and voluntary labelling, to guarantee the integrity of the EU high added-value as the “acorn Iberian pig ham”. The present study is focussed in transferring a database (470 samples) of IP tissue - analysed in a FOSS-NIRSystems 6500 (FNS6500) spectrometer, during the seasons 2009-2011 - to a portable/miniature instrument MicroNIR-Onsite, VIAVI (MN1700). A set of 30 samples of adipose tissue was taken from a slaughterhouse during 2015-2016, being analysed in parallel in the satellite (FNS 6500) and master (MN 1700) instruments. Latter on, they were divided in two sets: N = 10 for building the standardization matrices and N = 20 for the validation of the cloning procedure. The algorithm Piece-Wise Direct Standardization (PDS) was applied. The best standardisation matrix was applied to the library of 470 samples taken in the FNS 6500, enabling an excellent fitting between both instruments, as shown the RMSCs statistic calculated in the satellite before and after the standardization and in the master - 108457 vs 22519 vs 17646 μlog 1/R – and the GH distance before and after standardisation between both instruments 437.41 vs 2.06.

Listeria monocytogenes is the major etiologic agent for foodborne Listeriosis in humans from consumption of readyto- eat (RTE) food. According to Center for Disease Control and Prevention, an estimated 1,600 people contract Listeriosis each year with approximately 260 deaths. This high rate of mortality has alerted the Food Safety Inspection and Services to release the Notice 23-99, Instructions for Verifying the L. monocytogenes Reassessment, on August 3, 1999 for their inspectors. According to the FDA’s Bacterial Analysis Manual Chapter 10, L. monocytogenes in RTE food samples is detected via microbiological culture-based tests, qPCR, pulsed-field gel electrophoresis, and other alternative methods. Unfortunately, these methods are time consuming (48-72 hours) and require dedicated laboratory facility. Thus, to develop a real-time L. monocytogenes biosensor, we isolated L. monocytogenes specific oligopeptides displayed on bacteriophages using modified biopanning procedures. In order to account for major temperature dependent morphological alterations of L. monocytogenes at 4°C versus 37°C, we used bacterial cells adapted to either temperature as the target in our biopanning. To date, we have isolated several candidate probes that can recognize either cold-adapted, warm-adapted L. monocytogenes cells, or both types of bacterial cells. Our isolated probes will be used on the magnetoelastic biosensor platforms for real-time detection of L. monocytogenes in RTE foods stored at 4°C or in samples/fluids for bacterium adapted to human body temperature.

Foodborne illness is a common public health problem because food can be contaminated with pathogens at any point in the farm-to-table continuum. This paper presents a method of capturing a quantity of a specific bacterial pathogen in a large volume of liquid using a biomolecular recognition filter. The filter consists of support frames made of a soft magnetic material and solenoid coils for magnetization/demagnetization of the frames. This filter is a planar, multi-layered arrangement of strip-shaped, phage-immobilized magnetoelastic (ME) biosensors that are magnetically held and arrayed on the filter frames. As a large volume of liquid passes through the biomolecular filter, the pathogen of interest is captured by the phage immobilized ME biosensors. This biomolecular filter is designed to capture a specific pathogen and allow non-specific debris to pass, thus avoiding a common clogging issue in conventional bead filters. In this work, single layer, double layers and triple layers of filter were test to capture Salmonella Typhimurium in a large volume of water. The effects of multiplication of filter layers on Salmonella capture efficiency will be discussed.

The present work explores the possible utilization of hyperspectral devices, to evaluate olive fruit ripening in order to define optimal harvesting strategies and/or to perform an in depth characterization of the product addressed to “canned olive” productions, whose conservation characteristics, and related organoleptic attributes, strongly condition market price and producers revenue. A comparison was performed between two different hyperspectral sensing devices: the 1st one working at laboratory scale (Specim SisuCHEMA XL™: 1000-2500 nm) acquiring hyperspectral images and the 2nd one based on a portable architecture (ASD FieldSpec 4™ Standard-Res: 350-2500 nm) acquiring spectra on “spot” bases. Olive fruits collected spectra, acquired with the different sensing architectures, have been correlated with the maturity index and the harvesting time. To reach these goals a chemiometric approach, finalized to set up Partial Least Square (PLS) regression models able to predict olive fruits ripening and quality, was applied. Results have been compared in a proximity sensing perspective and in a “on-line” quality control logic, both finalized to maximize olive fruit derived product quality (i.e. olive oils and/or canned olives) in a costs/benefits perspective taking into account the different sensing architectures and their integrated utilization.

This study aimed to evaluate the possibility to utilize portable hyperspectral devices in order to perform fast, reliable, low cost and robust monitoring of kiwifruits ripening both in field and/or in processing storage facilities. This determination is of great interest, not only to assess kiwi taste according to market requirement, but also to set up efficient handling and conservation strategies avoiding loss of product. To reach these goals two different sensing units have been utilized, that is i) a MicroNIR™ JDSU Spectrometer working in the NIR wavelength range 950-1650 nm and ii) a ASD FieldSpec 4 ® Standard-Res field portable spectroradiometer working in the NIR wavelength range 350-2500 nm. Kiwifruits collected spectra have been correlated with classical quality indicators (i.e. Total Soluble Solid Content the Dry Matter content) following a chemiometric approach based on spectra preprocessing, data explorative analysis and Partial Least Square modelling. Kiwifruits ripening characteristics have been satisfactory predicted by both the units. Results have been compared in a costs/benefits perspective taking into account the different sensing architectures, their costs and easiness of use.

Fruits provide essential nutrition in most natural form suitable for human beings. They are best when ripened naturally. However, industrialization has provided many ways for quick ripening and for extended shelf life of fruits. Detection of artificial ripening could be done by sophisticated methods like chemical analysis in lab or visual inspection by experts, which may not be feasible all the time. Of all the fruits, banana is the most consumed fruit around the world. Adulteration of banana can have devastating effects on masses on scale. It is figured, bananas are potentially ripened using carcinogens like Calcium Carbide(CaC₂). In this paper, we propose and devise a novel and automatic method to classify the naturally and artificially ripened banana using spectral and RGB data. Our results show that using a Deep Learning (Neural Network) on RGB data, we achieve accuracy of up-to 90%.and using Random Forest and Multilayer Perceptron (MLP) feed forward Neural Network as classifiers on spectral data we can achieve accuracies of up-to 98.74% and 89.49% respectively.

Spectrometers are widely used for characterizing materials. Recently, filter-based spectrometers have been pro- posed to lower the manufacturing cost by replacing optical components with low-cost wavelength-selective filters, but at the expense of possibly lowered signal quality. We present compressive spectrometers which, based on the compressive sensing principle, are able to recover signal with improved quality from measurements acquired by a relatively small number of low-cost filters. We achieve high quality recovery by leveraging the fact that spectrometer measurements typically follow the shape of a smooth curve with a few spikes. We validate our method with real-world measurements, and release our dataset to facilitate future research.

In this paper imaging spectroscopy in the near-infrared wavelength range (900 nm-2500 nm) is evaluated for the inspection of peeled potato tubers. The high spectral resolution of the acquired data and the use of the near-infrared wavelength range make it possible to detect defects that are not visible to the human eye. The main inspection goal is the detection of potato tubers with defects that may influence the quality of French fries. The defects of interest are remains of peel, discoloration and water-, starch- or sugar distributions of the potato tubers. In this work we focus on the detection of increased sugar concentrations inside the potato which lead to an undesirable browning during the frying process. Multivariate statistical methods are used to discriminate between the pixel spectra for the defect regions and those for the remaining non-defect regions. The measurement setup and preliminary results of the data analysis are presented.