This PDF file contains the front matter associated with SPIE Proceedings Volume 11754, including the Title Page, Copyright information, and Table of Contents.

Near Infrared (NIR) Spectroscopy is a powerful technology which can be implemented as a non-destructive tool to make decisions related to cultural practices and harvesting. However, prior to the incorporation of NIR sensors at field level as an analytical technique, a routine analysis procedure should be established. In this sense, this research is focused on the development of a methodology based on the use of a portable NIR instrument to monitor the growth process and to establish the optimum harvest time of spinach plants in the field. For this aim, calibration models for dry matter and nitrate contents were developed by means of Partial Least Squares (PLS) regression, using one spectrum per plant for dry matter content and nine spectra per plant for nitrate content taken with a portable spectrophotometer MicroNIR™ OnSite-W (908– 1676 nm). After that, to set a routine analysis methodology, the validation of the models was carried out using a single spectrum per plant selected at random and the suitability of the predictions was measured considering the Hotelling’s T² statistic, whose control limit value was as inferior to 60. The results demonstrated that once the calibration models were developed, only one spectrum per plant will enable to predict successfully dry matter and nitrate contents. Therefore, the methodology established will allow to monitor spinach plants during their growth in the field based on internal quality and safety indexes.

Near infrared (NIR) spectroscopy can be a fast and reliable candidate for the non-destructive and in-situ classification of almonds by bitterness, when analysed in bulk. With that purpose, in-shell and shelled sweet and bitter almonds were analysed using a handheld diode array NIR spectrophotometer (950-1650 nm). Models were constructed using partial least squares-discriminant analysis (PLS-DA) and the optimum threshold value was set up using the Receiver Operating Characteristic (ROC) curves. The models correctly classified 95% of in-shell and 100 % of shelled samples belonging to the external validation sets. The excellent performances obtained for the classification models of the in-shell and shelled almonds analysed in bulk will enable to remove bitter almonds from the sweet almond batches and, with it, to avoid selling those batches containing bitter almonds that could lead to product depreciation.

Feeding dairy cows with Total Mixed Rations (TMR) is a cost-effective way to obtain high milk yield. Animal nutritionists are demanding accurate information on the main chemical constituents of TMR to properly feed lactating cows. The use of portable NIRS devices could provide an affordable answer. This work analysed a total of 121 TMR using two portable NIRS instruments for the prediction of dry matter, crude protein and neutral detergent fibre. The paper evaluated whether there were significant differences between the predictive capacities of the models developed from analytical data expressed “ as dry matter” or “ as is basis”.

Food authentication and quality checks can be carried out by applying machine learning algorithms on spectral data acquired from miniature spectrometers. This is a very appealing solution as the cost-effectiveness of miniature spectrometers extends the range of consumer electronics available for ordinary citizens in the fight against food fraud, widens the range of their applications and shortens the processing time for any in-situ scenario. In this paper, a study of olive oil purity and quality check feasibility carried out on spectral data acquired from a miniature spectrometer is presented. The aim is to gauge the ability of such a device to differentiate between pure olive oil from ones adulterated with vegetable oils on a relatively large dataset. The paper presents a pipeline encompassing various steps including data pre-processing, dimension reduction, classification, and regression analysis. That is, data collected from miniature spectrometers can be of low quality and exhibit distortions, high dimensionality, and collinearity. Hence, various filtering techniques including wavelets analysis, and normalisation algorithms including multiplicative scatter correction are used for pre-processing. Once the dimensionality is reduced using PCA, classical machine learning classification and regression analysis algorithms are deployed as part of a quality evaluation pipeline. This includes SVM, LDA, KNN, Random Forest and PLS. The obtained results show that very high rates of up to 98% can be achieved. An important consequence is that cost-effective miniature spectrometers augmented with a suitable machine learning component can attain comparable results obtained using non-portable and more expensive spectrometers.

The potential of near infrared hyperspectral imaging over the spectral range of 900 - 2500 nm was investigated for identification of aflatoxin contamination on corn kernels. A total of 600 kernels were used with 3 treatments, namely, 200 kernels inoculated with the AF13 fungus (aflatoxigenic), 200 kernels inoculated with the AF36 fungus (nonaflatoxigenic), and 200 kernels inoculated with sterile distilled water as control. One hundred kernels from each treatment were incubated at 30 °C for 5 and 8 days, separately, and then the kernels were dried and surface wiped to remove exterior signs of mold prior to imaging. The mean spectra including mean reflectance and absorbance, and the textural features consisting of contrast, correlation, energy and homogeneity, were extracted separately from the endosperm regions of single kernels. The partial least-squares discriminant analysis (PLS-DA) models were established using extracted mean spectra or textural features as individual inputs. The full spectral PLS-DA modeling results indicate that the mean spectra including both reflectance and absorbance spectra performed significantly better than using the textural features in identifying aflatoxin contamination on corn kernels. Using the mean reflectance and absorbance spectra between 925 and 2484 nm, the full spectral PLS-DA models achieved mean overall prediction accuracies of 88.3% and 86.3% when taking 20 ppb as the classification threshold. The corresponding means of overall prediction accuracies were 85.5% and 85.6% when 100 ppb was applied as the classification threshold. The extracted textural features were not found to be useful in identifying aflatoxin contamination.

Wheat (Triticum spp.) is a widely grown cereal crop and is one of the most important staple foods internationally. Grading is an imperative step in wheat production to ensure that the grains are of acceptable quality and safety for food processing and consumption. Current wheat grading practices are performed manually which is tedious, time-consuming, and subjective. This study aimed to investigate the feasibility of NIR hyperspectral imaging (HSI) and chemometrics to discriminate sound wheat from four common defects. Defective wheat included heat-damaged, Fusarium-damaged, sprout-damaged, and immature kernels. A wide variety of pre-processing techniques and classification algorithms (logistic regression, partial least squares-discriminant analysis, linear discriminant analysis, k-nearest neighbours, decision trees, random forests and support vector machines) were evaluated for both two-way and multiclass analyses. For the two-way classifications, accuracies between 98.6% and 100% were achieved for each defective category. The multiclass analyses had a decreased performance, where the best model attained an accuracy of 84.6%. Given that agricultural products vary significantly in inherent characteristics due to genetics, cultivation practices, handling and storage, the overall results achieved are highly successful. This study shows that HSI is capable of effectively discriminating sound wheat from common occurring defects, offering a promising alternative grading technique to the cereal industry.

The determination of the fatty acid profile in almonds has a huge interest to establish the nutritional value of the product. Hyperspectral Imaging (HSI) integrates both the spectral and spatial dimensions, enabling a rapid and non-destructive evaluation of the composition and distribution of quality indexes in agricultural products. The objective of this study was the determination of the two main unsaturated fatty acids -oleic and linoleic, in shelled almonds analysed in bulk using a HSI system working in the spectral range 946.6 to 1648.0 nm. The predictive models were developed using the mean spectrum extracted from the ROI of each sample and applying Partial Least Squares (PLS) regression. Subsequently, the external validation of the best models was carried out using the mean spectrum of each ROI and pixel-by-pixel. The results showed a good performance for the fatty acids analysed (R²cv = 0.78 and SECV = 2.17 for oleic and R²cv = 0.77 and SECV = 1.83 for linoleic), confirming the feasibility of using HSI as a non-destructive analytical tool to assess the lipid composition and its distribution in the almonds processed in bulk, as well as to include their nutritional properties in the labelling.

Microbial water quality monitoring is an essential component of food safety. E. coli bacterium is the major indicator organism used in assessing microbial water quality but dense sampling of water to assess spatial variability is impractical. The objective of this work was to test the hypothesis that sUAS imaging can provide information about the differences in E. coli habitats across two Maryland irrigation ponds and guide water sampling. We used modified GoPro cameras and a MicaSense RedEdge camera in flights shortly before sampling. Ponds P1 (0.37ha) and P2 (0.48ha) were sampled from a boat in the same locations, biweekly, during the 2018 growing season. Average concentrations of E. coli were 0.60±0.04 and 1.04±0.04 (mean ± st. error, log CFU/100 mL) in P1 and P2, respectively. The random forest (RF) machine learning algorithm was applied to relate ground sampling data with co-located image sections. The sensitivity of results to parameters of the RF algorithm was assessed with multiple scenarios. The most influential parameters for both ponds were maximum tree depth and minimum leaf size. The maximum R2 values in predictions of E. coli concentrations were 0.941 (0.943) and 0.532(0.565) in training and validation datasets, respectively, for pond P1 (P2). The most influential inputs for both ponds were red, blue, and green obtained after demosaicing images in the visible range, while P1 included red and blue obtained after demosaicing infrared images. Overall, accurate estimation of E. coli concentrations from imagery data is possible and benefits from tuning algorithm control parameters.

We report on an Unmanned Aerial Vehicle (UAV) agricultural data acquisition system to substantially improve how farming is currently practiced. The demonstrated innovative platform simplifies data acquisition with high spatial and temporal resolution and high accuracy with affordable cost and hence accessible for wider communities. Our goal was to create a product capable of analyzing multiple crucial soil parameters via one unit and transfer the accumulated preprocessed data to an airborne mobile subsystem. The system consists of two modules working in harmony. The Soil Data Monitoring Probe (SDMP) is a stationary unit housing various soil probes, whereas the Airborne Data Acquisition System (ADAS) is a mobile unit that can be placed on a UAV. The SDMP captures soil metrics, preprocesses them and stores the data to an SD card, to be delivered using a NRF24 transceiver. The unit itself is battery and solar-powered, regulated power is fed through a custom-designed motherboard to the Arduino mega microcontroller (ATMEGA-2560), and internal/external modules. We estimated that, with its low-power design and the complementary solar power, it can work months without interruption. The ADAS is lightweight and was mounted on a drone. It initiates data collection by interrogating sleeping SDMP’s based on geolocation stamps by waking them up from deep sleep mode with an interrupt. The ADAS is supported by a custom motherboard to support a raspberry pi zero with Wi-Fi capabilities. We demonstrated the operational prototype system with 2-meter spatial resolution.

Quartz crystal microbalance (QCM) sensors have been applied to detect foodborne pathogens such as Salmonella Typhimurium, E.coli O157:H7, and Campylobacter jejuni. As pathogens are placed on vibrating quartz surface, the change in mass of the pathogens affects the characteristic of a QCM. The presence of pathogens that antibody captures can be correlated to the shift in resonance frequency. Thus, theoretical description is necessary to understand the relationship between the change in frequency and mass. In this work, the relationship between theoretical and experimental results is examined by comparing the frequency shift caused by different type of liquids. In general, a QCM can be represented by a Butterworth-Van-Dyke (BVD) circuit made up of resistance R, inductance L, and capacitance C. With physical properties of quartz, viscosity-density product of the liquid has an effect on inductance as well as resistance. As a preliminary experiment, measurements of mixtures of water and glycerol were conducted to evaluate results from the different levels of viscosity and density. The results of the experiments showed that higher viscosity and density resulted in lower resonant frequencies. With regard to theoretical calculation, increase of R and L resulted in a proportional increase in the square root of the viscosity-density product. Increased lumped parameters explains the decreased resonant frequency. Therefore, the shift of the resonant frequency of the load and unloaded QCM could be calculated based on the admittance from circuit components. Blank (air) sample, water, glycerol and water mixture have shown proportional shift in the resonant frequencies. The experiments and theoretical model were matched within reasonable range. The average difference between the theory and the experiments (Matlab/FEM model) was 7.04 %.

Problems with “clenbuterol”, a synthetic β₂-adrenergic stimulant that includes clenbuterol, ractopamine, and albuterol, have been increasing in recent years. This article focuses on the detection of albuterol, chemically called 3,6-dihydroxy1-methyl-5-oxo-3,5-dihydro-2H-indolium betaine. It is difficult to directly measure the content of clenbuterol in pork due to its messy and uneven composition, which requires a destructive pre-treatment process that takes time and effort. In this study, a method for subsurface food inspection was presented based on a newly developed line-scan spatially scattering Raman spectroscopy technique. A spatially offset Raman spectroscopy system based on line-scan Raman chemical imaging system was built, which was used to collect spatially offset Raman spectra from the samples. Under the condition of one CCD exposure, the spatially offset Raman spectroscopy system can collect a series of Raman spectra simultaneously in a narrow space interval and a wide offset range. Through data analysis, the rare clenbuterol species in the sample can be determined by the position of characteristic peaks. The system was used to collect salbutamol Raman signal, and the results showed that the characteristic peaks of salbutamol were consistent with the standard characteristic peaks, so it was possible to use the system to detect lean meat in meat.

The increasing normative requirements and market competitiveness lead the agricultural sector and the food industry to constantly look for new fast and non-destructive classification logics that can be applied for product sorting applications and/or quality control actions. With reference to hazelnut production, the dried fruits must be sorted from unwanted foreign bodies or inedible hazelnuts that can negatively affect the quality of the final product. In this perspective, the utilization of HyperSpectral Imaging (HSI) can be applied to set-up a novel hazelnuts quality control. Hazelnuts and contaminants were acquired by a push-broom hyperspectral device working in the Short-Wave InfraRed (SWIR: 1000-2500 nm) region. A PLSDA model was set up in order to identify 3 classes of products (i.e. edible hazelnuts, hazelnut shells and rotten hazelnuts) with the highest level of efficiency in full spectrum mode (Precision = 0.92, Accuracy = 0.94, Efficiency = 0.94). Subsequently, different variable selection methods (i.e. Interval PLSDA, Selectivity Ratio and Variable Importance in Projection score methods) were adopted in order to identify the fundamental bands to recognize the 3 classes and evaluate which of the variable selection methods shows efficiency values close to the values obtained by the full spectrum mode. VIP score-based classification showed the best performance, with Precision, Accuracy and Efficiency values equal to those based on full spectrum PLSDA. Classification results suggest that this methodological approach can be powerful to develop and implement hazelnut sorting and/or quality control strategies. Moreover, the variable selection approach allows to increase processing speed , compared to that in full spectrum mode, making possible online applications directly at plant scale.