Search

Advances in optics technology and computational processing have brought multispectral and hyperspectral imaging to commercial sorting of fruits and vegetables, yet the application of imaging to single cereal seeds has lagged due to the enormity in numbers of seeds and challenges posed by lighting, shadowing, and seed curvature that are less problematic with larger objects. This study examined the effect of region of interest (ROI) size on the seed surface with respect to the ability to sort seed into accept and reject categories. Regions of interest (ROI) size ranged from 5 centrally located pixels arranged in a cross to all pixels (typically 100) contained in the viewed surface of a kernel. Two modeling structures were used; the first involving all 87 samples, with approximately 220 kernels per sample, in which mixture level of sound and fusarium-damaged kernels is known, but individual kernel class is unknown; and the second involving 5 samples, of which an equal number of 287 known sound and known fusarium-damaged kernels were used. Accordingly, the larger set model characterised the dispersion (by standard deviation) of kernel-to-kernel reflectance at a single representative wavelength, while the smaller set was used to develop linear discriminant analysis classification models using one to three wavelengths. With either case, it was found that the smaller ROIs should be sufficient for a two-class (accepts, rejects) structure.

This study is directed towards developing a fast, non-destructive, and easy-to-use handheld multimode spectroscopic system for fish quality assessment. We apply data fusion of visible near infra-red (VIS-NIR) and short wave infra-red (SWIR) reflectance and fluorescence (FL) spectroscopy data features to classify fish from fresh to spoiled condition. Farmed Atlantic and wild coho and chinook salmon and sablefish fillets were measured. Three hundred measurement points on each of four fillets were taken every two days over 14 days for a total of 8400 measurements for each spectral mode. Multiple machine learning techniques including principal component analysis, self-organized maps, linear and quadratic discriminant analyses, k-nearest neighbors, random forest, support vector machine, and linear regression, as well as ensemble and majority voting methods, were used to explore spectroscopy data measured on fillets and to train classification models to predict freshness. Our results show that multi-mode spectroscopy achieves 95% accuracy, improving the accuracies of the FL, VIS-NIR and SWIR single-mode spectroscopies by 26, 10 and 9%, respectively. We conclude that multi-mode spectroscopy and data fusion analysis has the potential to accurately assess freshness and predict shelf life for fish fillets and recommend this study be expanded to a larger number of species in the future.

Relative to standard red/green/blue (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 light emitting diodes (LEDs) to collect images in multiple wavelengths. This paper proposes a new methodology to support the design of a hypothesized system that uses three imaging modes—fluorescence, visible/near-infrared (VNIR) reflectance, and shortwave infrared (SWIR) reflectance—to capture narrow-band spectral data at only three to seven narrow wavelengths. Simulated annealing is applied to identify the optimal wavelengths for sparse spectral measurement with a cost function based on the accuracy provided by a weighted k-nearest neighbors (WKNN) classifier, a common and relatively robust machine learning classifier. Two separate classification approaches are presented, the first using a multi-layer perceptron (MLP) artificial neural network trained on sparse data from the three individual spectra and the second using a fusion of the data from all three spectra. The results are compared with those from four alternative classifiers based on common machine learning algorithms. To validate the proposed methodology, reflectance and fluorescence spectra in these three spectroscopic modes were collected from fish fillets and used to classify the fillets by species. Accuracies determined from the two classification approaches are compared with benchmark values derived by training the classifiers with the full resolution spectral data. The results of the single-layer classification study show accuracies ranging from ~68% for SWIR reflectance to ~90% for fluorescence with just seven wavelengths. The results of the fusion classification study show accuracies of about 95% with seven wavelengths and more than 90% even with just three wavelengths. Reducing the number of required wavelengths facilitates the creation of rapid and cost-effective spectral imaging systems that can be used for widespread analysis in food monitoring/food fraud, agricultural, and biomedical applications.

Plant phenomics has been rapidly advancing over the past few years. This advancement is attributed to the increased innovation and availability of new technologies which can enable the high-throughput phenotyping of complex plant traits. The application of artificial intelligence in various domains of science has also grown exponentially in recent years. Notably, the computer vision, machine learning, and deep learning aspects of artificial intelligence have been successfully integrated into non-invasive imaging techniques. This integration is gradually improving the efficiency of data collection and analysis through the application of machine and deep learning for robust image analysis. In addition, artificial intelligence has fostered the development of software and tools applied in field phenotyping for data collection and management. These include open-source devices and tools which are enabling community driven research and data-sharing, thereby availing the large amounts of data required for the accurate study of phenotypes. This paper reviews more than one hundred current state-of-the-art papers concerning AI-applied plant phenotyping published between 2010 and 2020. It provides an overview of current phenotyping technologies and the ongoing integration of artificial intelligence into plant phenotyping. Lastly, the limitations of the current approaches/methods and future directions are discussed.

Methods that combine targeted techniques and chemometrics for analyzing food authenticity can only facilitate the detection of predefined or known adulterants, while unknown adulterants cannot be detected using such methods. Therefore, the non-targeted detection of adulterants in food products is currently in great demand. In this study, FT-IR and FT-NIR spectroscopic techniques were used in combination with non-targeted chemometric approaches, such as one-class partial least squares (OCPLS) and data-driven soft independent modeling of class analogy (DD-SIMCA), to detect adulterants in almond powder adulterated with apricot and peanut powders. The reflectance spectra of 100 pure almond powder samples from two different varieties (50 each) were collected to develop a calibration model based on each spectroscopic technique, each model was then evaluated for four independent sets of two varieties of almond powder samples adulterated with different concentrations of apricot and peanut powders. Classification using both techniques was highly sensitive, the OCPLS approach yielded 90&ndash, 100% accuracy in different varieties of samples with both spectroscopic techniques, and the DD-SIMCA approach achieved the highest accuracy of 100% when used in combination with FT-IR in all validation sets. Moreover, DD-SIMCA, combined with FT-NIR, achieved a detection accuracy between 91% and 100% for the different validation sets and the misclassified samples belong to the 5% and 7% adulteration sets. These results suggest that spectroscopic techniques, combined with one-class classifiers, can be used effectively in the high-throughput screening of potential adulterants in almond powder.

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.

Aflatoxins, commonly found in corn and corn-derived products, can cause severe illness in animals and humans if consumed in significant amounts. Early detection is critical to preventing illness, but the most sensitive and effective of commonly used screening tools for aflatoxins are expensive and cumbersome methods based on chromatography or imunoassays that require technical expertise to perform. Multiple hyperspectral imaging techniques, including reflectance in the visible and near-infrared (VNIR) region and short-wave infrared (SWIR) region, fluorescence by 365 nm ultraviolet (UV) excitation, and Raman by 785 nm laser excitation, were used for detection of aflatoxin in ground maize. Four classification models based on linear discriminant analysis (LDA), linear support vector machines (LSVM), quadratic discriminant analysis (QDA), and quadratic support vector machines (QSVM) algorithms were developed for classification with each hyperspectral imaging mode. The multivariate classification models in combination with different preprocessing methods were applied for screening of maize samples naturally contaminated with aflatoxin. The classification accuracies for fluorescence with QSVM, VNIR with QSVM, SWIR with LSVM, and Raman with LSVM were 95.7%, 82.6%, 95.7%, and 87.0%, respectively, with no false-negative error at the cutoff of 10 μg/kg. The SWIR and fluorescence models showed slightly higher performance accuracies, suggesting that they may be more effective and efficient analytical tools for aflatoxin analysis in maize compared to conventional wet-chemical methods. These methods show promise as inexpensive, and easy-to-use screening tools for food safety, to rapidly detect aflatoxins in maize or other food ingredients intended for animal or human consumption.

Rapid and reliable inspection of food is essential to ensure food safety, particularly in mass production and processing environments. Many studies have focused on spectral imaging for poultry inspection, however, no research has explored the use of multispectral fluorescence imaging (MFI) for on-line poultry inspection. In this study, the feasibility of MFI for on-line detection of fecal matter from the ceca, colon, duodenum, and small intestine of poultry carcasses was investigated for the first time. A multispectral line-scan fluorescence imaging system was integrated with a commercial poultry conveying system, and the images of chicken carcasses with fecal contaminants were scanned at processing line speeds of one, three, and five birds per second. To develop an optimal detection and classification algorithm to distinguish upper and lower feces-contaminated parts from skin, the principal component analysis (PCA) and partial least square discriminant analysis (PLS-DA) were first performed using the spectral data of the selected regions, and then applied in spatial domain to visualize the feces-contaminated area based on binary images. Our results demonstrated that for the spectral data analysis, both the PCA and PLS-DA can distinguish the high and low feces-contaminated area from normal skin, however, the PCA analysis based on selected band ratio images (F630 nm/F600 nm) exhibited better visualization and discrimination of feces-contaminated area, compared with the PLS-DA-based developed chemical images. A color image analysis using histogram equalization, sharpening, median filter, and threshold value (1) demonstrated 78% accuracy. Thus, the MFI system can be developed utilizing the two band ratios for on-line implementation for the effective detection of fecal contamination on chicken carcasses.

Substitution of high-priced fish species with inexpensive alternatives and mislabeling frozen-thawed fish fillets as fresh are two important fraudulent practices of concern in the seafood industry. This study aimed to develop multimode hyperspectral imaging techniques to detect substitution and mislabeling of fish fillets. Line-scan hyperspectral images were acquired from fish fillets in four modes, including reflectance in visible and near-infrared (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. All fillet samples were DNA tested to authenticate the species. A total of 24 machine learning classifiers in six categories (i.e., decision trees, discriminant analysis, Naive Bayes classifiers, support vector machines, k-nearest neighbor classifiers, and ensemble classifiers) were used for fish species and freshness classifications using four types of spectral data in three different datasets (i.e., full spectra, first ten components of principal component analysis, and bands selected by 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 the overall best performance for both species and freshness inspection, and it will be further investigated as a rapid technique for detection of fish fillet substitution and mislabeling.

Cases of imitation or fake food materials are sometimes produced and sold for purposes of economic fraud. However, while some imitation or fake food materials merely incorporate lower quality or cheaper alternative ingredients that are safe to eat, others fakes are produced using non-edible or hazardous ingredients that are unsafe for consumption. The latter group includes fake eggs which are often difficult to identify by eye. Such fakes have been found in various parts of Asia, made from harmful ingredients such as sodium alginate, tartrazine dye, gypsum powder, and paraffin wax. The objective of this study is to evaluate the use of Raman spectral analysis for nondestructive, noninvasive identification of fake eggs. In this study, fake eggs were prepared and then Raman spectroscopic and imaging data were collected from both the fake eggs and real chicken eggs. Classification of the fake and real eggs was tested using both Raman spectroscopy (1800–600 cm−1) with multivariate analysis methods and Raman hyperspectral imaging (1500–390 cm−1) with waveband optimization. The results demonstrated that both techniques are able to differentiate fake eggs from real eggs.

Viability is an important quality factor influencing seed germination and crop yield. Current seed-viability testing methods rely on conventional manual inspections, which use destructive, labor-intensive and time-consuming measurements. The aim of this study is to distinguish between viable and nonviable soybean seeds, using a near-infrared (NIR) hyperspectral imaging (HSI) technique in a rapid and nondestructive manner. The data extracted from the NIR&ndash, HSI of viable and nonviable soybean seeds were analyzed using a partial least-squares discrimination analysis (PLS-DA) technique for classifying the viable and nonviable soybean seeds. Variable importance in projection (VIP) was used as a waveband selection method to develop a multispectral imaging model. Initially, the spectral profile of each pixel in the soybean seed images was subjected to PLS-DA analysis, which yielded a reasonable classification accuracy, however, the pixel-based classification method was not successful for high accuracy detection for nonviable seeds. Another viability detection method was then investigated: a kernel image threshold method with an optimum-detection-rate strategy. The kernel-based classification of seeds showed over 95% accuracy even when using only seven optimal wavebands selected through VIP. The results show that the proposed multispectral NIR imaging method is an effective and accurate nondestructive technique for the discrimination of soybean seed viability.

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.

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.