This PDF file contains the front matter associated with SPIE Proceedings Volume 11016, including the Title Page, Copyright information, Table of Contents, Author and Conference Committee lists.

Turmeric powder (Curcuma longa L.) is known for its use in foods, in medicine, and as a cosmetic. In recent years, economically driven contamination of turmeric powder with different chemicals is increasing. This study used a 1064 nm hyperspectral Raman imaging system for detection of Sudan Red G dye contamination in turmeric powder. Sudan Red was mixed with turmeric powder at five concentration levels (1%, 5%, 10%, 15%, and 20%- w/w). Each mixture sample was packed in a sample container. A Raman chemical image of each sample was acquired across the 7.5 mm x 7.5 mm surface area using a 0.25 mm step size. The spectral fingerprint of turmeric and Sudan Red were identified and used to obtain a binary image from the Raman chemical image of each sample. A simple threshold method was applied to convert the contaminant pixels into white pixels and turmeric pixels into the black (background) pixels. The detected Sudan Red pixels were correlated with the actual concentration in the sample. The result shows that the Sudan Red pixels in the sample image is linearly correlated (R² = 0.99) with the actual concentration of the sample. This study demonstrated the 1064 nm hyperspectral Raman imaging system as a potential tool to detect chemical contaminants in turmeric powder.

The potential of near infrared (NIR) hyperspectral imaging over the 900-2500 nm spectral range was examined for discrimination of artificially-inoculated corn kernels with aflatoxigenic and non-aflatoxigenic strains of Aspergillus flavus in this study. The two A. flavus strains, aflatoxigenic AF13 and non-aflatoxigenic AF36 were used for inoculation on corn kernels. Four treatments were included, with each treatment consisting of 100 kernels. Each treatment of 100 kernels were artificially inoculated with AF13 or AF36 strain and incubated at 30 °C for 3 and 8 days, separately. The mean spectra were extracted from the collected NIR hyperspectral images for individual corn kernels, and then based on the mean spectra, the principal component analysis combined with linear discriminant analysis (PCA-LDA) method was employed to establish the classification models. The pairwise classification models were established by the PCA-LDA method to discriminate the AF36-inoculated and the AF13-inoculated kernels at different incubation days. All the overall accuracies obtained by the pairwise models were ≥98.0%. A common model that takes the AF13-inoculated kernels at different incubation days as one class and the AF36-inoculated kernels at different incubation days as the second class, achieved an overall accuracy of 99.0% for the prediction samples. This indicates a great potential of using NIR hyperspectral imaging to classify corn kernels infected by aflatoxigenic and non-aflatoxigenic A. flavus regardless of infection time.

Commercial Raman systems generally conduct imaging and spectroscopy measurements at subcentimeter scales. Such small spatial ranges cannot be used to inspect food samples with large surface areas (e.g., tomato fruit and beef steak), which is not convenient for food experiments. A line-scan macro-scale Raman system has been developed using a 785 nm line laser to implement high-throughput Raman chemical imaging (RCI) for food safety and quality research. A one-axis positioning table is used to move the samples to accumulate hyperspectral data using a pushbroom method. A dispersive Raman spectrograph is used in the system, which can be configured to backscattering RCI mode for surface inspection and spatially offset Raman spectroscopy (SORS) mode for subsurface inspection. In-house developed LabVIEW software is used to fulfill functions for system control, hardware parameterization, and data transfer. The systems is flexible and versatile for food test, and it has been used to evaluate safety and quality of various food and agricultural products, such as detecting chemical adulterants mixed in food powders, mapping carotenoid content on carrot cross section, imaging whole surface of pork shoulder, and authenticating foods and ingredients through packages.

This study is concerned with development of multi-sensor data fusion technology to detect woody breast fillets in the poultry industry. The common practice in commercial plants for detection of woody breast fillets is through subjective evaluations of various visual traits although the woody breast myopathy is uniquely distinguished with tactile attributes of muscle hardness and rigidity of fillets. This study extends and improves the previously developed rapid and non-invasive 2D machine vision technique that measures muscle rigidity using a single 2D camera to detect woody breast fillets moving on conveyor system. This 2D machine vision technology is currently under development for commercialization. In this study, multi-sensor data fusion of 2D and 3D shapes and color features is proposed to further improve the performance of the single camera-based technology. A preliminary study found that information fusion of different physical properties such as muscle rigidity, muscle out-bulging shape, and presence of hemorrhagic lesions on the skin-side surface of the fillets could improve the detection accuracy than that provided by individual sensors.

Our goal is to analyze spectral imaging data using multiple optical imaging instruments available in USDA-ARS and SafetySpect laboratories to provide analysis along three axes of classification of fish fillets: 1. farm-raised vs. wild-caught species; 2. fresh vs. frozen fillets; 3. Species A vs. Species B targeting most mislabeled fish types in the US market. We are collecting spectral signatures using four imaging systems: (1) Reflectance spectral imaging in the visible and NIR (400-1000 nm), (2) Reflectance spectral imaging in the short wave infra-red (SWIR) (1000-2500 nm), (3) Fluorescence spectral imaging with UVA and violet illumination, (4) Raman imaging with a 785 nm laser excitation. The fish fillet samples were purchased from online vendors. We image with each of the modalities and then freeze, thaw and reimage each fillet (2 cycles of freeze/thaw) to demonstrate effect of freeze/thaw process in the multimode spectral signatures. All fish fillet samples are DNA tested to ensure the species marketed are not mislabeled. We use feature extraction/selection strategy for different modes of measurements based on the measurement physics and biological/chemical characteristics. We analyze different combinations of feature extraction and selection techniques and operate an exhaustive search, optimization, and fusion to find out the most important features using different imaging modes. This process helps identify which imaging mode (or combination) will have the highest impact and yield 95%+ classification accuracy. This optimization procedure will be based on cost function (sensitivity, specificity, area under the curve) from receiver operating characteristics (ROC) curve.

Bacterial processes during shelf life of meat are of paramount importance for packaging, preservation and conservation of food. As those processes are linked with fermentation and production of some gaseous species, a sealed thermostatic test cell has been developed and fitted with optical gas sensor as a tool for microbiology and food researchers. The gas sensing instrument is able to detect ammonia and water vapour through a fiber coupled, Herriott multipass absorption cell used with a Distributed Feedback (DFB) laser operating around 1514nm in a Tunable Diode Laser Absorption Spectroscopy (TDLAS) setup. A software defined Wavelength Modulation Spectroscopy (WMS) was used together with a custom made fitting routine to discriminate ammonia and water vapour. A Non Dispersive Infrared (NDIR) instrument at 4260nm is also included in the gas path for carbon dioxide sensing. Headspace gas is sampled from the cooled thermostatic cell and circulated with a diaphragm pump through the sensing instrument while measurements are logged. The whole device consisting of electro optical sensors assemblies and circulation fittings has been integrated into a portable unit. Validation tests were performed at different temperatures with calibrated solutions of ammonia and water directly placed in the thermostatic cell. The results were compared with a reference electrochemical sensor and show that the multipass ammonia sensor is able to reach a resolution in the order of 100ppb with a time response of less than one minute.

Arcobacter (formerly classified as Campylobacter spp.) are curved-to helical, Gram-negative, aerobic/microaerobic bacteria increasingly recognized as human and animal pathogens. In collaboration with Lincoln and Purdue University, we report the first experimental result of laser-based classification method of bacterial colonies of these species. This technology is based on elastic light scatter (ELS) phenomena where incident laser interacts with the whole volume of the colony and generates a unique fingerprint laser pattern. Here we report a novel development and application of deep learning algorithm to classify the scatter patterns of Arcobacter species using variational autoencoders (VAE). VAE creates set of normal distributions. Each of these distributions are responsible for certain properties of the original images. We used VAE to identify features in the features space for several hundred images which includes size of the colony based on scatter size, intensity of the image, and, the number of rings within the image, and so on. Thus each sample within our image database can be coded with sets of features that facilitates fast preliminary search for similar images allowing clustering of similar patterns in feature space. In addition, such initial selection could assist in identifying non-bacterial scatter patterns (i.e. bubbles or dust spots in the agar), or doublets where two colonies are overlapping during the acquisition time thus removing non-biological artifacts prior to analysis. An interesting result was that while VAE created far more realistic synthetic images closer to the original image, a simple autonencoder resulted in better cluster separation.

Laser-induced breakdown spectroscopy (LIBS) is a technique developed in the last few decades for simultaneous multi-element characterization of various materials. Multiplexed detection of analytes is particularly useful in the realm of food contaminant detection, where the contaminant can be one or a combination of adulterants. Paper-based assays are an emerging platform for food-contaminant detection. However, most paper-based assays do not perform multiplexed detection. For food contaminant outbreak prevention and remediation, rapid multiplexed detection could make a difference in response speed. This study applies LIBS to the concept of multi-analyte detection on paper-based bioassays. In the envisioned bioassay, a variety of analytes are labeled with unique lanthanides, a technique common to the well-established field of mass cytometry. The presence of single or multiple lanthanide labels indicates the presence of single or multiple types of contaminants. We aim to implement LIBS for multiplexed detection of lanthanide labels. To investigate data analysis approaches for multi-lanthanide detection, we evaluate univariate data analysis and spectral unmixing approaches on samples containing combinations of europium, dysprosium, gadolinium, praseodymium, and neodymium. We find that the intense signal generated by Eu, matrix effects, selfabsorption, and spectral overlap affect the outcome of the results. Future studies will continue the investigation to identify the most appropriate approach.

This paper presents a portable magnetoelastic (ME) biosensor system that enables rapid, on-site detection of pathogenic bacteria. The system utilizes a patented portable resonant frequency analyzer and two 1 mm long ME biosensors (biosensors coated with and without phage binding a specific pathogen) enabling real-time measurement of resonant frequency changes. By comparing the response of the biosensors, the presence of the specific pathogen can be detected. In this work, detection of Salmonella Typhimurium cells was demonstrated, and it was found that down to 2,500 cfu can be detected in less than 10 min. The detection limit can be improved by using a smaller sensor (e.g., 500 um long sensors) and an optimal chamber design increasing the probability of bacterial cells striking the biosensor surface.

Outbreaks of foodborne illness due to pathogenic bacteria have been identified worldwide and have been associated with the consumption of contaminated agricultural products. The main objective of this research is to develop a rapid method for pathogen detection using Raman spectroscopy (RS). Direct detection in culture media and surface-enhanced Raman scattering (SERS) were used to identify Escherichia coli, Escherichia coli O157:H7, Salmonella spp., Listeria monocytogenes, Staphylococcus aureus, Bacillus cereus, and Bacillus thuringiensis. Bacterial isolates were cultured on selective media for 24 h at 37°C or 30°C and then tested with RS. A portable 785 nm point-scan Raman system was developed at ARS USDA for this purpose and multiple laser current and exposure times were tested to establish optimal conditions. Seven nanoparticles and three substrates were evaluated for optimal bacterial detection using label-free SERS. Raman peaks were very weak in direct detection and the bacteria were not identified using direct or SERS approaches. However, two gold nanoparticles consistently showed SERS peaks at 878.9, 1086, and 1455 cm^-1 and relative differences in Raman intensity were observed among each of the tested bacteria. This method can be used to lay a foundation for future research such as SERS combined with chemometric analysis and label-based SERS approaches.

Acorn Iberian ham (Jamón Ibérico de Bellota) is one of the most expensive luxury foodstuffs produced in Europe, with a highly appreciated smell and flavour. Its recognized high-sensorial quality and health properties are mainly due to the traditional outdoor feeding system (Montanera) of Iberian pigs (IP), which provides high standards of animal welfare. Nowadays, one of the frauds affecting this product is the use of “special compound feeds” to simulate the fat composition of the acorns through the inclusion of sources of oleic acid like the ones found in pigs fed outdoors. The high prices paid for a cured leg of Iberian ham –ranging from hundreds to thousands of euros- leads to many opportunities for mislabelling and fraud. Fatty acid content of the adipose tissue could provide evidence of the feeding system. Gas chromatography (GC) is used at industry level for production control purposes. However, it is costly and time-consuming, and it is only applied to batches of animals rather than individual pigs. The main goal of this study was to use spectra belonging to a portable NIRS instrument (MicroNIR Onsite Lite, Viavi Solutions Inc.) for on–site quantitative (fatty acid content) analysis of individual Iberian pork carcasses at the slaughterhouse. Performance of this portable instrument was compared with an at-line NIRS monochromator. PLS models were built and optimized resulting in standard errors of cross validation ranging from 0.83 to 0.84 for palmitic acid, 0.94 to 0.99 for stearic acid, 1.47 to 1.56 for oleic acid and 0.53 to 0.58 for linoleic acid.

Microbial quality of irrigation water is the public health issue that is the subject of regulatory actions mandated by the Food Safety Modification Act. Concentrations of the bacterium E. coli are currently used to derive the microbial water quality metrics. Direct E. coli monitoring requires substantial resources. We hypothesized that drone based imagery can reflect fine-scale differences in E. coli habitats and its survival in irrigation ponds. We tested this assumption using the DJI Matrice 600 Pro sUAS equipped with modified GoPro’s and a MicaSense camera. Digital numbers from imagery were averaged across the 46 sampling locations and compared to 10 water quality parameters using rule-based machine-learning algorithms for estimating E. coli concentrations at a Maryland irrigation pond. Cross-validation with Bootstrap obtained statistical distributions of RMSE and determination coefficient R2 of the decision rule based estimators. The average R2 was 0.79 which is comparable with R2 of estimates from the full set of water quality parameters. Overall, the results indicate the promise of proximal sensing with drone-based imagery to serve as an information source for evaluating microbial water quality.

Luminescence based detection has been widely used in diverse science and engineering applications. The recent development of the smartphone has enabled end users to utilize this communication device as a portable detector and instruments such as a microscope, fluorimeter, colorimeter, and spectrometer. To transform the smartphone into a bioluminescence detector, our group developed an advanced signal processing algorithm and an optical chamber designed for efficient photon capture. This solution was required to overcome the typical sensitivity of the CMOS-based smartphone camera such that sub-nano to pico Watt levels of power can be measured with conventional smartphones. Preliminary experiments conducted with the bioluminescent Pseudomonas fluorescens M3A shows a detection limit of approximately 10⁶ CFU/ml. To achieve sensitive detection while maintaining the portability, we explored using the recently developed silicon photomultiplier (SiPM), and designed a portable bioluminescence sensor which shows a 2-3 order higher sensitivity on calibration sample testing. Finally, for live sample testing, Escherichia coli O157:H7 was inoculated on a ground beef sample and subjected to luminescence phage based detection and a luminescence signal was generated from the bacteriophage infection and detected within 8-10 h after spiking.

Hyperspectral image technology is a powerful tool, but oftentimes the data dimension of hyperspectral images must be reduced for practical purposes, depending on the target and environment. For detecting defects on a variety of apple cultivars, this study used hyperspectral data spanning the visible (400 nm) to near-infrared (1000 nm). This paper presents the preliminary results from the selection of optimal spectral bands within that region, using a sequential feature selection method. The selected bands are used for multispectral detection of apple defects by a classification model developed using support vector machine (SVM). As a result, five optimal wavelengths were selected as key features. When using optimal wavelengths, the accuracy of the SVM and SVM with RBF kernel achieved accuracies over 90% for both the calibration and validation data set. However, the results of SVM with RBF kernel (>80%) based on image was more robust than SVM model (>50%). Moreover, SVM with RBF model classified between bruise and sound regions as well specular. The result from this study showed the feasibility of developing a rapid multispectral imaging system based on key wavelengths.

In recent years, the quality and safety issues related to flour and pasta products have attracted great attention from the society. The quality of flour will directly affect the quality of downstream pasta products, as well as the physical and mental health and economic benefits of consumers. In this study, the illegal additive benzoyl peroxide in flour was the research object, and the rapid real-time non-destructive detection of benzoyl peroxide in flour was realized by Raman hyperspectral technique. By comparing the Raman spectra of pure benzoyl peroxide and pure flour, several Raman spectral characteristic peaks of benzoyl peroxide and their assignments were found. Characteristic peaks with strong signal at 1001 cm^-1 and 1777 cm^-1 were extracted for quantitative analysis. A gradient concentration of benzoyl peroxide-doped flour samples from 1% to 0.05% was prepared. And a series of pretreatment including S-G 5-point smoothing and background removal were performed to extract the number of effective benzoyl peroxide pixels in the mixed sample. And the proportion of benzoyl peroxide pixel points in total pixel points with different benzoyl peroxide concentrations was acquired. By comparing the relationship between the proportion and the concentration of benzoyl peroxide, a quantitative analysis model for the benzoyl peroxide doping in flour was established. The verification results show that there was good correlation between the proportion and the concentration of benzoyl peroxide. Both the averaged benzoyl peroxide signal intensities of effective pixel points and the number of effective pixels were combined for quantitative analysis. The research provided a methodological support for the detection of additives in flour by hyperspectral techniques and was a reference for the detection of dopants in food.

A rapid and reliable inspection technique for determining sanitation status of produce processing equipment surfaces in processing facilities is needed to help reduce potential food safety risks. In this study, fluorescence imaging methods were evaluated to detect produce residues on the surfaces of food processing equipment such as stainless steel (STS). Contamination spots on the STS were created using droplets of a range of dilutions of carrot juice. Hyperspectral fluorescence images of the sample surfaces were obtained using excitation light sources based on ultraviolet LEDs (365 nm) and on violet LEDs (405 nm) for comparison. Image and spectral data were analyzed to determine optimal single bands and two-band ratios to detect the juice residue spots of the STS, and a support vector machine (SVM) algorithm was applied to the ratio images to determine classification accuracies. These results suggest that the simple multispectral fluorescence imaging methods can potentially be incorporated into portable imaging devices for spot-checking food contact surfaces for contaminants in processing facilities.