This PDF file contains the front matter associated with SPIE Proceedings Volume 11421, including the Title Page, Copyright information, and Table of Contents.

A device based on non-dispersive infrared spectroscopy has been developed for leak testing through non-invasive evaluation of escaping carbon dioxide (CO₂) outside of Modified-Atmosphere-Packed (MAP) cheese containers. The targeted samples are bags of processed mozzarella sealed in modified atmosphere at the end of the manufacturing process, in which the internal CO₂ concentration is higher than 10% vol. for product shelf-life extension. The device performs in-line measurement on moving samples on a conveyor belt with the application of a calibrated test pressure to the samples to stimulate leaks through any possible package defect such as holes, cuts, wrinkles and swelling in the seal. The sensor device takes advantage of a multichannel suction manifold and an array of high-speed, optical non- dispersive CO₂ sensors for a space-resolved sensing of any leaked carbon dioxide around the sample perimeter. Any variation in the detected carbon dioxide level when compared with ambient background is correlated with a faulty package. Moreover, the device is able to report and log an approximate fault location and automatically remove the defective package from the production line.

This study developed multimode hyperspectral imaging techniques to detect substitution and mislabeling of fish fillets. Line-scan hyperspectral images were collected from fish fillets in four modes, including reflectance in visible and nearinfrared (VNIR) region, fluorescence by 365 nm UV excitation, reflectance in short-wave infrared (SWIR) region, and Raman by 785 nm laser excitation. Fish fillets of six species (i.e., red snapper, vermilion snapper, Malabar snapper, summer flounder, white bass, and tilapia) were used for species differentiation and frozen-thawed red snapper fillets were used for freshness evaluation. A total of 24 machine learning classifiers were used for fish species and freshness classifications using four types of spectral data in three different subsets (i.e., full spectra, first ten components of principal component analysis, and bands selected by a sequential feature selection method). The highest accuracies were achieved at 100% using full VNIR reflectance spectra for the species classification and 99.9% using full SWIR reflectance spectra for the freshness classification. The VNIR reflectance mode gave an overall best performance for both species and freshness inspection.

A very low cost multispectral detector is developed and bench marked with full spectrometer measurements by measuring internal quality parameters of kiwis. The multispectator detector uses self-referenced reflectance to reduce measurement variations. It is demonstrated that even when using only twelve wavelengths, only a small loss of accuracy occurs with respect to a spectrometer in measurements of solid soluble content and dry matter. Further, using classification, similar accuracy is achieved in placing the fruits in bins based on their quality parameters. The measurements are rapid (less than 5 seconds), non-destructive and the system costs less than $50.

Gradient Temperature Raman spectroscopy (GTRS) applies the temperature gradients utilized in differential scanning calorimetry (DSC) to Raman spectroscopy, providing a straightforward technique to identify molecular rearrangements that occur at or near phase transitions. 20 Mb threedimensional data arrays with 1.0 or 0.2°C increments allow complete assignment of solid, liquid and transition state vibrational modes, including low intensity/frequency vibrations that cannot be readily analyzed with conventional Raman. We compared GTRS and DSC data for commercial fish oil supplements that are excellent sources of docosahexaenoic acid (DHA; 22:6n-3) and eicosapentaenoic acid (EPA; 20:5n-3). Krill oils and whole fish and shellfish have nearly all their PUFA contained within phospholipids (PL). Development of any fast-throughput optical identification system for seafood products will require improved PL Raman data. We also analyzed molecularly hydrated PL with DHA and other unsaturated lipids. Each fish oil and PL have a unique, distinctive response to the thermal gradient, which graphically and spectroscopically differentiates them. The entire set or any subset of the any of the contour plots, first derivatives or second derivatives can be utilized to create a graphical standard to quickly authenticate a given product or source material.

The potential of line-scan hyperspectral Raman imaging system equipped with a 785 nm line laser was examined for discrimination of healthy, AF36-inoculated and AF13-inoculated corn kernels in this study. The AF36 and AF13 strains were used as representatives for the aflatoxigenic and non-aflatoxigenic A. flavus fungal varieties. A total of 300 kernels were used with 3 treatments, namely, 100 kernels inoculated with the AF13 fungus, 100 kernels inoculated with the AF36 fungus, and 100 kernels inoculated with sterile distilled water as control. The kernels were all incubated at 30 °C for 8 days and then dried and surface wiped to remove exterior signs of mold. The kernels were imaged from endosperm side over the wavenumber range of 103-2831 cm^-1. The mean spectrum was extracted from the Raman image of each kernel, and preprocessed with adaptive iteratively reweighted penalized least squares, Savitzky-Golay smoothing and min-max normalization. Based upon the preprocessed group mean spectra, a total of 35 local Raman peaks were identified. With the spectral variables at the identified local peak locations as inputs of discriminant models, the 3-class principal component analysis-linear discriminant analysis (PCA-LDA) models ran 20 random times, achieved a mean overall prediction accuracy of 91.13% along with a standard deviation value of 3.36%.

Raman spectroscopy is commonly used in analytical chemistry for identification of molecules based on the analysis of spectral lines in the Raman spectrum corresponding to molecular vibrations. For solids and gases, usually high spectral resolution is required. For many other applications, however, a high spectral resolution is less critical, while compactness and reduced system cost are increasingly important. Several studies have already shown the applicability of low resolution Raman spectroscopy (LRRS) systems to a wide variety of fields, such as in situ monitoring of process control, on-site detection of illicit drugs and explosives, on-site detection of water contamination, and safety analysis of edible oil. In this work, we illustrate how imec’s CMOS based hyperspectral (HS) filter technology can be used to build a very compact, low cost, mass-manufacturable Raman spectrometer where spectral resolution can be traded for increased sensitivity, hence better SNR and/or shorter acquisition time. Furthermore, unlike other types of Raman spectrometers, the proposed system can be tailored to the application by only targeting specific Raman bands, e.g., 400-1500cm^-1 and 3000-4000cm^-1, by selecting the right set of HS filters for a given wavelength. This enables further improvements in system performance, or even use for Raman imaging. The HS filter technology is currently already used in various commercially available HS cameras for imaging applications in the VIS-NIR (470-900nm), NIR (600-970nm) and SWIR range (1100-1700nm) to study the reflectance or transmission spectra of the imaged targets. Proof-of-concept Raman measurements have been successfully performed in a laboratory setup where the impact of using different HS filter selections or layouts and different cooled and uncooled cameras can be tested. Furthermore, the measurements closely match the predicted spectra obtained with our in-house developed HS-based Raman spectrometer simulation software, which can be used for design space exploration to optimize each of the system components (optical module, rejection filters, HS filters and image sensor) for the target application.

The research published on animal protein by-products (ABPs) has been conducted using at- line instruments. The aim of this study is to evaluate different strategies to transfer a large spectral database of ABPs recorded in a monochromator instrument, to a FT-NIR instrument coupled to a fibre optic probe of 100 metres length, for the on-site quality control. The results obtained demonstrated that, once a large spectral data base of ABPs (more than 1300 samples) has been transferred from the monochromator to the FT-NIR instrument, the calibrations developed for on-site analysis have similar accuracy that those used previously for at-line analysis.

In this study, NIR hyperspectral imaging and multivariate data analysis were used to classify game meat species, namely Springbok (Antidorcas marsupialis) and Blesbok (Damaliscus pygargus phillipsi). The animals (6 blesbok and 6 springbok) were harvested in Witsand and Elandsberg in the Western Cape, South Africa. Longissimus thoracis et lumborum (LTL) muscles were excised and left to bloom for ca. 30 minutes prior to imaging. Thereafter, the moisture was wiped off the surface to avoid specular reflectance. NIR hyperspectral images were collected with a linescan system (HySpex SWIR-384) in the spectral range 950 – 2500 nm. Data were pre-processed with Savitzky-Golay smoothing and derivatives (2nd order polynomial, 2nd derivative, 15 point smoothing) and noisy regions in the spectra, specifically between 1884.9nm -2500nm, were removed. In addition, two data analysis methods, the pixel and object wise approach, were evaluated. In the object wise approach the muscles were segmented into ca 2 cm ROI’s, of which the mean was computed. For both pixel and object wise approaches, there was no distinct separation of the species with PCA. When PLS-DA models were developed, the object wise approach proved to be superior with a classification accuracy of 96%, whereas that of the pixel wise approach was 62%. It is evident that NIR hyperspectral imaging can be used to distinguish between the two species, with the object wise being the optimal option of the two approaches, as it represents the mean spectra of each object.

Conventional methods for the determination of chemical parameters of the fruit like soluble solids and acid content are often complicated and destructive, cannot be run on a large scale and are still far away from being implemented to large volumes of products or even better to individual piece fruits. In this study, the potential of hyperspectral imaging was evaluated for quantifying solid soluble content (SSC) and titratable acidity (TA) in intact oranges. Hyperspectral images (900–1700 nm) of 264 oranges collected during 2017 and 2018 at different maturation stages in Southern Spain farms were recorded. Partial least-squares analysis (PLS), Artificial Neural Network (ANN), optimized Support Vector Machine (SVM) and Gaussian Process Regression (GPR), as well as different spectral pre-processing methods, were tested for their effectiveness in quantifying titratable acidity (TA) and solid soluble content (SSC) in intact oranges. Random samples were chosen to validate the models by cross-validation. The best-selected models were then applied to a validation set of “unknown” samples and standard errors of prediction as well as correlation coefficients between actual and predicted values were calculated. Finally, a prediction map was developed to display the concentration distribution of the TA and SSC in the orange fruit, demonstrating that hyperspectral imaging (HSI) technique was feasible to quantify parameters in citrus fruit and can be further used for monitoring the quality of oranges at pre- and post-harvest in real-time.

The objective of this study was to evaluate the effect of the cutting method on the quality of the dried ginger (Zingiber officinale Rosc.) by NIR hyperspectral imaging and computer vision systems. The cutting method of ginger was done vertically (slices) and horizontally (splits). The mean spectra were extracted from the collected NIR hyperspectral images (950-1,655 nm) for individual ginger samples and the partial least squares regression (PLSR) was employed to establish the prediction models. Determination coefficient (R²) of PLSR models based on non-pretreatment spectra for predicting moisture contents and rehydration rates of ginger samples were 0.960 and 0.957, respectively. The prediction maps of ginger slices and splits showed the same dehydration and rehydration patterns, which the moisture contents and rehydration rates in center parts were higher than edges. However, the shrinkage rates of ginger slices were higher than splits, while rehydration rates of ginger splits were higher than slices.

Currently, it is very demanded by nutritionists the availability of real-time on farm analysis for Total Mixed rations ( TMR ) quality control at the level of individual dairy farms. This study refers to the prediction of Crude Protein ( CP) in TMR, after transference of a library file ( N =394 ) of TMR samples from a monochoromator instrument, to two on-farm portable instruments (NIR4Farm, AUNIR, UK and AURORA, GraiNIT, Italy). The results obtained demonstrated that CP can be predicted by NIRS at “ on farm level”, with an accuracy similar to the most expensive at-line laboratory instruments.

Soil is a precious, essentially non-renewable, resource presently endangered by human activities. The road to protection goes through knowledge: there is the need to raise the global understanding of the importance of soil. Therefore, diffuse measurement of soil properties is central to soil conservation and management. We present a device tailored to monitor the soil quality by adopting a low-cost open-source multi-sensor approach. The device measures in the field soil temperature, moisture, density and pH. Furthermore, the device has a penetrometer to obtain a depth-resolved soil hardness characterization down to 60cm from the surface. The device is controlled via Bluetooth by a custom-built Android App. The application georeferences each data and uploads the generated _les to a server for data elaboration aimed to build and populate a map with all the collected soil quality measurements. The device is equipped with a temperature/moisture sensor, a 0-5kg load cell used as weight scale for density measurements, a reflectance spectra sensor for automatic reading pH test strips and, on the bottom side, a penetrometric tip mounted on a 100kg load cell coupled with an optical time-of-flight sensor for depth-referenced penetrometer measurements. Preliminary test evidenced good correlation between in-the-field observations and traditional laboratory tests. The device is easily also by unexperienced users, allowing for applications both in agronomic and in environmental fields, ranging from educational purposes and ICT learning of soil sciences to scientific projects related to soil monitoring and awareness raising Citizen Science initiatives.

The increasing demand of the horticultural sector in terms of quality and safety assurance stresses the need of the producers and the agri-food industry of implementing non-destructive analysis techniques. Near infrared spectroscopy (NIRS) has proven to be an increasingly practical option for satisfying this demand. Recently a new generation of NIRS instruments has been developed, being necessary their previous evaluation before their incorporation for quality and safety assurance along the food supply chain. For this purpose, 230 summer squashes, grown outdoors in the province of Cordoba (Spain), were analyzed to determine quality (dry matter content (DMC) and soluble solid content (SSC)) and safety (nitrate content) parameters using two spectrophotometers, MicroNIR^TM Pro 1700 and Matrix-F, ideally suited for the in situ and online analysis, respectively. A linear calibration strategy - modified partial least squares regression, MPLS - were used for the development of predictive models. The results obtained showed NIRS technology, by means of new generation sensors, is a potential tool for the non-destructive measurement of DMC (RPD_cv = 1.76 and RPD_cv = 1.98), SSC (RPD_cv = 1.62 and RPD_cv = 1.63) and nitrate content (RPD_cv = 1.77 and RPD_cv = 1.36), for the MicroNIR^TM Pro 1700 and Matrix-F, respectively. This would enable to improve the quality and safety control of this vegetable throughout the whole supply chain, i.e. in field and in the processing plant.

Our goal is to develop a reliable and cost-effective spectral imaging system with sparse spectral measurements. Relative to standard RGB imaging systems, hyperspectral imaging systems offer superior capabilities but tend to be expensive and complex, requiring either a mechanically complex push-broom line scanning method, a tunable filter, or a large set of LEDs to collect images in multiple wavelengths. We would like to overcome these limitations by employing a novel spectral reconstruction algorithm to recreate the full-resolution reflectance or fluorescence spectrum from an optimized selection of images at a sparse set of wavelengths. This algorithm is aided by a single full-resolution spectrometer measurement representing an average value over the selected spatial scene. We use a genetic algorithm-based methodology to identify the optimal wavelengths for sparse spectral measurement and invoke a cost function that includes a weight vector to emphasize minimization of errors in key portions of the spectrum. To validate the proposed algorithm, reflectance spectra in the visible and NIR (400-1000 nm) and fluorescence spectra with UV illumination were collected from fish fillets to validate our methods. In this paper, we discuss the reconstruction algorithm and the genetic algorithm-based optimization method that we use to determine the optimal set of wavelengths for imagery collection. We also present results from a fish species classification study using the reconstructed spectra as feature sets for four common machine learning algorithms. The classification accuracies based on these reconstructed spectra are on par with the accuracies that result from using the original full spectral resolution data.

A machine learning algorithm combining reinforcement learning and supervised learning is demonstrated for training of near infrared spectroscopy data for non-destructive measurement of fruit quality. The model optimizes the combination of pretreatment methods, discriminant methods and calibration methods and also the parameters used in the methods to achieve highest prediction correlations. The model achieves better results than manual combinations of the previously demonstrated models.

Our goal is to use multiple spectroscopy methods in a single system and develop novel multimode spectroscopic data fusion techniques for fish species identification in real-time. We collected spectral signatures of fish fillets from six fish species using four hyperspectral imaging systems: (1) Reflectance spectral imaging in the visible and NIR (VIS-NIR), (2) Reflectance spectral imaging in the short wave infrared (SWIR), (3) Fluorescence visible spectral imaging with UVA and violet excitation, (4) Raman imaging with a 785 nm laser excitation. All fish fillet samples were confirmed by DNA testing. We built multiple classification/ dimension reduction combination methods to calculate the average sensitivity and associated variance for each class and each spectroscopy mode. In our prototype, the derived statistics are used to form policies for Monte Carlo prediction reinforcement learning. We compared the results of our weighted fusion decisions against individual spectroscopy mode decisions to show an overall sensitivity improvement. We believe this is the first reported use of reinforcement learning applied to multimode spectroscopy data classification in food fraud applications.

Hyperspectral imaging is a powerful remote sensing tool capable of capturing rich spectral and spatial information. Although the origins of hyperspectral imaging are in terrestrial remote sensing, new applications are emerging rapidly. Owing to its non-destructive nature, hyperspectral imaging has become a useful tool for monitoring produce ripeness. This paper describes the process that uses a visible near-infrared (VNIR) hyperspectral imager from Resonon, Inc., coupled with machine learning algorithms to assess the ripeness of various pieces of produce. The images were converted to reflectance across a spectral range of 387.12 nm to 1023.5 nm, with a spectral resolution of 2.12 nm. A convolutional neural network was used to perform age classification for potatoes, bananas, and green peppers. Additionally, a genetic algorithm was used to determine the wavelengths carrying the most useful information for age classification. Experiments were run using RGB images, full spectrum hyperspectral images, and the genetic algorithm feature selection method. Results showed that the genetic algorithm-based feature selection method outperforms RGB images for all tested produce, outperforms hyperspectral imagery for bananas, and matches hyperspectral imagery performance for green peppers. This feature selection method is being used to develop a low-cost multi-spectral imager for use in monitoring produce in grocery stores.

A new method for detection both external and internal quality attributes of tomato was proposed in this paper. The comprehensive quality detection could be completed by the image and spectra analyses based on the optical sensing system. The images processing method contained three steps: (1) morphological filtering; (2) binarization and (3) circle fitting. The first step was applied to reduce the random noises in the raw images. The second step was aimed to obtain the binary images that contained the contour information of the tomato edge. And the circle fitting algorithm was used to obtain the final diameter information of the tomato samples. The values of R and the RMSE for size prediction results of the tomato samples were 0.9813 and 1.269 mm, respectively. For the spectra analysis, the light scatter effects, including addition coefficient and multiplication coefficient in the raw spectra, were the main reasons for the calibration failure of the multivariate linear model such as PLS, MLR and PCR. Thus, the NSR method was used to eliminate the light scatter effects in this paper. Compared with the other method, NSR method was advantages in higher prediction performance and the simpler calculation. The RMSEP values of the final PLS model were 0.2936 % and 2.0129 a.u. for the SSC and a*, respectively. Thus, the optical sensing system combined with effective information processing method was able to detect the tomato external and internal quality attributes, which could be more suitable to apply in the food processing enterprise in the practice.

Mastitis is an infectious-contagious disease that causes an inflammation of the udder that affects a high proportion of dairy cows throughout the world. The difficulty of its diagnosis (which requires culture media especially for its isolation) and inefficiency of antibiotics in your treatment, what become a fearsome enemy if detected its presence in a dairy farm, since only very rigorous hygiene and disposal measures of the positive cows, are the measures known for controlling it. While it is true that subclinical mastitis does not usually increase in greatly the amount of colony forming unit (CFU)/ml (x1000) of tank milk, can contribute some bacteria potentially harmful to human health, also alters the composition of milk. This research introduces the development of an analytical methodology for on-site monitoring the CFU/ml in raw milk at farm level by using a portable NIR sensor MicroPhazir^TM NIR spectrometer, using a total of 1266 liquid milk samples, scanned at room temperature without pre-treatment. Samples were divided into two sub-sets. The training set composed of 1197 samples, and a set of 69 samples to external validation. Classification models were used for the prediction of CFU/ml in milk at legal level: < 400 and ≥ 400 CFU/ml(x1000), achieving less than 3% of penalties.

The citrus sector is one of the most dynamic and important agricultural sectors. For the international market, it is of great interest the estimation of crop yield prior to harvest, since this yield estimation at the immature green stage could influence the future market price and allow producers to plan the harvest in advance. The aim of this work was to stablish the first steps to set up a methodology for the selection of the relevant bands to distinguish between green oranges and leaves and to detect external defects, which will allow citrus yield to be estimated on tree. Images were acquired from oranges and leaves from an orchard in Jeju island (Jeju, Republic of Korea), using a hyperspectral reflectance imaging system working in the range 400–1000 nm. Analysis of variance (ANOVA) and principal component analysis (PCA) were used to select the main wavelengths for this purpose; next, a band ratio coupled with a simple thresholding method was applied. The system correctly classified over the 90% of the pixels for both objectives, confirming that it is possible to use just few wavelengths to estimate harvest yield in oranges, although further studies are needed for the application of this system in the field, where other factors must be taken into account, such as sun-light illumination, shadows, etc. Therefore, this research can be considered as a preliminary step for designing a multispectral system capable of being mounted on unmanned aerial vehicles (UAVs) to estimate orange yield and defects.