Potential Of Near-infrared Hyperspectral Imaging Research Articles

Using computational learning for non-melanoma skin cancer and actinic keratosis near-infrared hyperspectral signature classification

BackgroundThe early detection of Non-Melanoma Skin Cancer (NMSC) is essential to ensure patients receive the most effective treatment. Diagnostic screening tools for NMSC are crucial due to high confusion rates with other types of skin lesions, such as Actinic Keratosis. Nevertheless, current means of diagnosing and screening patients rely on either visual criteria, that are often conditioned by subjectivity and experience, or highly invasive, slow, and costly methods, such as histological diagnoses. From this, the objectives of the present study are to test if classification accuracies improve in the Near-Infrared region of the electromagnetic spectrum, as opposed to previous research in shorter wavelengths. MethodsThis study utilizes near-infrared hyperspectral imaging, within the range of 900.6 and 1454.8 nm. Images were captured for a total of 125 patients, including 66 patients with Basal Cell Carcinoma, 42 with cutaneous Squamous Cell Carcinoma, and 17 with Actinic Keratosis, to differentiate between healthy and unhealthy skin lesions. A combination of hybrid convolutional neural networks (for feature extraction) and support vector machine algorithms (as a final activation layer) was employed for analysis. In addition, we test whether transfer learning is feasible from networks trained on shorter wavelengths of the electromagnetic spectrum. ResultsThe implemented method achieved a general accuracy of over 80 %, with some tasks reaching over 90 %. F1 scores were also found to generally be over the optimal threshold of 0.8. The best results were obtained when detecting Actinic Keratosis, however differentiation between the two types of malignant lesions was often noted to be more difficult. These results demonstrate the potential of near-infrared hyperspectral imaging combined with advanced machine learning techniques in distinguishing NMSC from other skin lesions. Transfer learning was unsuccessful in improving the training of these algorithms. ConclusionsWe have shown that the Near-Infrared region of the electromagnetic spectrum is highly useful for the identification and study of non-melanoma type skin lesions. While the results are promising, further research is required to develop more robust algorithms that can minimize the impact of noise in these datasets before clinical application is feasible.

Nondestructive Detection of Codling Moth Infestation in Apples Using Pixel-Based NIR Hyperspectral Imaging with Machine Learning and Feature Selection.

Codling moth (CM) (Cydia pomonella L.), a devastating pest, creates a serious issue for apple production and marketing in apple-producing countries. Therefore, effective nondestructive early detection of external and internal defects in CM-infested apples could remarkably prevent postharvest losses and improve the quality of the final product. In this study, near-infrared (NIR) hyperspectral reflectance imaging in the wavelength range of 900–1700 nm was applied to detect CM infestation at the pixel level for three organic apple cultivars, namely Gala, Fuji and Granny Smith. An effective region of interest (ROI) acquisition procedure along with different machine learning and data processing methods were used to build robust and high accuracy classification models. Optimal wavelength selection was implemented using sequential stepwise selection methods to build multispectral imaging models for fast and effective classification purposes. The results showed that the infested and healthy samples were classified at pixel level with up to 97.4% total accuracy for validation dataset using a gradient tree boosting (GTB) ensemble classifier, among others. The feature selection algorithm obtained a maximum accuracy of 91.6% with only 22 selected wavelengths. These findings indicate the high potential of NIR hyperspectral imaging (HSI) in detecting and classifying latent CM infestation in apples of different cultivars.

Near-infrared hyperspectral imaging for polymer particle size estimation

This study examines the potential of near-infrared hyperspectral imaging for assessing the size of polymer particles in model fractions based on the scattering phenomena. Different fractions of ground polymers, either polymethyl methacrylate or polypropylene, were characterized by near-infrared spectra collected between 900 and 1700 nm. The possibility to estimate the size of polymer particles using hyperspectral images was confronted with a basic single spot near-infrared measurement. Hyperspectral imaging, in addition to the standard spectral data dimension, provides information about the spatial distribution of sample components and reveals changes in physical properties. Therefore, one can gain a better insight into the scattering phenomena and study the physical inhomogeneity of a sample in terms of particle size distribution. The partial least-squares models constructed to estimate particle size of polymers that were characterized by hyperspectral images (a pixel-based approach) outperforms models built for mean spectra regardless of the considered powdered polymer.

Detection of chocolate powder adulteration with peanut using near-infrared hyperspectral imaging and Multivariate Curve Resolution

This study aims to detect peanut flour adulteration in chocolate powder using near-infrared (NIR) hyperspectral imaging. Fifteen samples were prepared by mixing both food products in different proportions (0%, 0.1%, 1%, 10% and 100% of peanut) and measured using the hyperspectral camera. A preliminary Principal Component Analysis (PCA) was performed to investigate the structure of the data. Next, the Multivariate Curve Resolution - Alternating Least Squares (MCR-ALS) chemometric method combined with a selectivity constraint in the concentration matrix was applied to untangle the spectral data into a set of components representative of the main constituents found in the samples. Moreover, a detection algorithm based on the calculation of the Mahalanobis distance for every pixel to the model distribution of chocolate powder was implemented. This analysis revealed the complexity of the unmixing problem, allegedly due to the spectral signature overlap in the pixel field of view and because the pure products presented similar spectral signatures. MCR-ALS results were improved after the application of a selectivity constraint, which resulted in a higher performance of the detection algorithm. MCR-ALS detected from 0% to 2.2% of adulterated pixels in mixed samples. On the other hand, the selectivity-constrained MCR-ALS method provided detections from 0.03% to 17.0% in those samples. This pipeline showed that peanut adulteration can be detected even for the lowest concentration level tested (0.1% of peanut). This work highlights the potential of NIR hyperspectral imaging combined with chemometrics for detection purposes.

Nondestructive quality assessment of chili peppers using near-infrared hyperspectral imaging combined with multivariate analysis

There is an increasing demand of chili peppers due to their special taste and numerous applications through a number of markets, and high quality is crucial for both producers and customers. This research was aimed to investigate the potential of near-infrared hyperspectral imaging (HSI) for nondestructive quality assessment of chili peppers. Near-infrared HSI in the spectral range of 975–1646 nm was employed to acquire hyperspectral reflectance images of chili peppers. High-performance liquid chromatography and freeze-drying methods were conducted to obtain the reference values of capsaicinoid concentrations and water contents, respectively. Three different variable selection methods with successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS) and genetic algorithm-partial least squares (GA-PLS) were performed to remove the redundant information and select the optimal wavelengths. Quantitative models including partial least squares (PLS), extreme learning machine (ELM) and least-squares support vector machine (LS-SVM) were then developed to predict the capsaicinoid concentrations and the water content. The results show that the ELM models combined with the SPA method yielded the best prediction performances for the capsaicin and dihydrocapsaicin concentrations, and the water content, with the highest correlation coefficients of prediction (RP) of 0.83, 0.80 and 0.93, respectively. Distribution maps of capsaicin and dihydrocapsaicin concentrations for intact and cut chili peppers were obtained. Finally, classification models for discriminating pungent and non-pungent chili peppers with a classification accuracy of 98.0% were developed. The results demonstrate that near-infrared HSI technique is promising for pepper quality assessment.

Online detection and quantification of particles of ergot bodies in cereal flour using near-infrared hyperspectral imaging

ABSTRACTThe objective of this study is to assess near-infrared (NIR) hyperspectral imaging for the detection of ergot bodies at the particle level in cereal flour. For this study, ground ergot body samples and wheat flour samples as well as mixtures of both from 100 to 500,000 mg kg–1 were analysed. Partial least squares discriminant analysis (PLS-DA) models were developed and applied to spectral images in order to detect the ergot body particles. Ergot was detected in 100% of samples spiked at more than 10,000 mg kg–1 and no false-positives were obtained with non-contaminated samples. A correlation of 0.99 was calculated between the reference values and the values predicted by the PLS-DA model. For the cereal flours containing less than 10,000 mg kg–1 of ergot, it was possible for some samples spiked as low as 100 mg kg–1 to detect enough pixels of ergot to conclude that the sample was contaminated. However, some samples were under- or overestimated, which can be explained by the lack of homogeneity in relation to the sampling issue and the thickness of the sample. This study has demonstrated the potential of NIR hyperspectral imaging combined with chemometrics as an alternative solution for discriminating ergot body particles from cereal flour.

Assessment of pesticide coating on cereal seeds by near infrared hyperspectral imaging

Classical chromatographic methods, such as ultra performance liquid chromatography (UPLC), are used as reference methods to assess seed quality and homogeneous pesticide coating of seeds. These methods have some important drawbacks since they are time consuming, expensive, destructive and require a substantial amount of solvent, among others. Near infrared (NIR) spectroscopy seems to be an interesting alternative technique for the determination of the quality of seed treatment and avoids most of these drawbacks. The objective of this study was to assess the quality of pesticide coating treatment by near infrared hyperspectral imaging (NIR-HSI) by analysing, on a seed-by-seed basis, several seeds simultaneously in comparison to NIR spectroscopy and UPLC as the reference method. To achieve this goal, discrimination—partial least squares discriminant analysis (PLS-DA)—models and regression—partial least squares (PLS)—models were developed. The results obtained by NIR-HSI are compared to the results obtained with NIR spectroscopy and UPLC instruments. This study has shown the potential of NIR hyperspectral imaging to assess the quality/homogeneity of the pesticide coating on seeds.

Detection of cucumber green mottle mosaic virus-infected watermelon seeds using a near-infrared (NIR) hyperspectral imaging system: Application to seeds of the “Sambok Honey” cultivar

The cucurbit diseases caused by cucumber green mottle mosaic virus (CGMMV) have led to a serious problem to growers and seed producers because it is difficult to prevent spreading through pathogen-infected seeds. Conventional detection methods for infected seeds such as biological, serological, and molecular measurements are not practical for measuring entire samples due to their destructive nature, and time, and cost issues. For this reason, it is necessary to develop a rapid and non-destructive novel technique for detecting seeds infestation. A near-infrared (NIR) hyperspectral imaging system was used to discriminate virus-infected seeds from healthy seeds with partial least square discriminant analysis (PLS-DA) and least square support vector machine (LS-SVM). The classification accuracy for virus-infected watermelon seeds were 83.3% with the best model, demonstrating the potential of NIR hyperspectral imaging for detection of virus-infected watermelon seeds.

Rapid and non-invasive quantification of intramuscular fat content of intact pork cuts

Having acquired near infrared (NIR) hyperspectral images of intact pork loin samples through an NIR hyperspectral imaging system, the efficiency of a variety of image processing techniques including texture pattern analysis techniques were applied to process hyperspectral images so as to determine the intramuscular fat (IMF) content non-destructively. After the segmentation of region of interest (ROI), the raw spectral, texture-based spectral and textural characteristics of pork images were extracted by spectral averaging and pattern recognition techniques namely Gabor filter and improved gray level co-occurrence matrix (GLCM), respectively. First derivatives of the non-filtered and the Gabor filtered spectra were also investigated. Full waveband partial least squares regression (PLSR) was employed to determine the optimal parameters of Gabor filter and GLCM, and to select optimal wavelengths for IMF prediction. A stepwise procedure was applied to the optimal wavelengths to further optimize them to key wavelengths. Multiple linear regression (MLR) models were built based on the key wavelengths. Mean spectra and the Gabor filtered spectra outperformed GLCM. The best result, represented by correlation coefficients of calibration (Rc), cross validation (Rcv) and prediction (Rp) of 0.89, 0.89, and 0.86, respectively, was achieved using the first derivative of Gabor filtered spectra at 1193 and 1217 nm. To visualize the IMF content in pork, the distribution maps of IMF content in pork were drawn using a mean spectra-based MLR model. These promising results highlight the great potential of NIR hyperspectral imaging for non-destructive prediction of IMF content of intact pork.

Feasibility Study on the Use of Near-Infrared Hyperspectral Imaging for the Screening of Anthocyanins in Intact Grapes during Ripening

The potential of near-infrared hyperspectral imaging to determine anthocyanins in intact grape has been evaluated. The hyperspectral images of intact grapes during ripening were recorded using a near-infrared hyperspectral imaging covering the spectral range between 900 and 1700 nm. Reference values of anthocyanins were obtained by HPLC-DAD. A number of spectral pretreatments and different mask development strategies were studied. Calibrations were performed by modified partial least-squares regression (MPLS) and present a good potential (RSQ of 0.86 and SEP values of 2.62 and 3.05 mg g(-1) of grape skin for nonacylated and total anthocyanins, respectively) for a fast and reasonably inexpensive screening of these compounds in intact grapes.

WATER AND FOOD QUALITY

Despite recent advances in food product development (e.g. nutraceuticals and functional foods) our understanding of the role of water in foods is still in its infancy. Monitoring of food quality via optical methods such as near infrared (NIR) and terahertz spectroscopy is made possible due to the interaction of electromagnetic energy and the water matrix. Aquaphotomics provides a framework for understanding the role of water in food systems and processes. This paper highlights the importance of water in food quality and presents a number of case studies that demonstrate the potential of NIR hyperspectral imaging for monitoring food quality through observation of water absorbance bands.

NIR Hyperspectral Imaging for Plastics Classification

Introduction N ear infrared (NIR) hyperspectral imaging (HSI) has been widely used in the pharmaceutical and food industries for quality monitoring and process control. This spread in applications has been mainly due to the development of instrumentation, computational ability in handling large amounts of data and data processing methods that can extract relevant information from the images. Hyperspectral images are threedimensional images in which a complete spectrum is measured for each spatial pixel. When the spectrum is recorded in the NIR region, an NIR hyperspectral image is obtained. Partial Least Squares-Discriminant Analysis (PLS-DA) is one of the most commonly used data analysis methods for classification of samples on the basis of conventional spectroscopy. When NIR hyperspectral images of different samples are recorded, PLS-DA may be applied to the spectral profiles of the different samples in order to classify them. As usual, a previous calibration step involving known samples is necessary so that PLS-DA is able to classify unknown samples. Plastic classification is an important field in which NIR HSI could be applied; correct identification and classification of plastics would enhance recycling efficiencies. Recently, some attempts at plastic classification by optical methods such as NIR, Raman spectroscopy, Fourier transform infrared (FT-IR) or differential scanning calorimetry (DSC) have been reported, but the existing literature on HSI-based methods is limited. HSI offers the important added benefit of being able to spatially map the location of various different plastic objects in a scene on the basis of their spectral signatures. In this study, we considered the potential of NIR HSI combined with PLS-DA for the classification of different types of polymers. Materials and methods The polymers studied were low-density polyethylene (LDPE), high-density polyethylene (HDPE), polypropylene (PP) and polystyrene PS. Three images have been analysed, denoted sample 1, sample 2 and sample 3. In Figure 1, a false colour image of sample 1 is shown. Samples 2 and 3 can be seen in Figure 2. Each of the images has two sides, the left side consists of a piece of each polymer type (individual polymers) and the right side consists of a mixture of seven pieces of each polymer (polymer mixture). In two of the images (samples 2 and 3), some of the mixed pieces of polymers are overlapped whereas in the other image (sample 1) the pieces are mixed but not overlapped. The initial dimensions of each image were 320 rows × 580 columns × 121 bands, where the first two dimensions (320 × 580) refer to the pixel or spatial dimension of the surface and the last dimension (121) refers to the number of wavelengths of each spectrum measured at each of the pixels. The NIR wavelength interval recorded spanned the range from 880 nm to 1720 nm with a resolution of 7 nm. PLS-DA is widely-used and known for sample classification. In this application, it consisted of a PLS2-based regression in which the response variables were categorical, expressing the class membership of the statistical units. It is a supervised method and it requires an initial calibration step involving some samples belonging to each known class. 2

Fungal Detection in Wheat Using Near-Infrared Hyperspectral Imaging

Different species of fungi infect grain in the field and storage facilities. Contamination by fungi in grain is detected and quantified by traditional methods, such as microbial incubation and microscopic detection, which are subjective, labor intensive, and time consuming. An accurate and timely detection technique for fungal growth in grain is needed to prevent grain from spoiling and to reduce quality loss. In this study, the potential of near-infrared hyperspectral imaging to detect fungal infection in wheat was investigated. Wheat kernels infected with storage fungi, namely Penicillium spp., Aspergillus glaucus, and Aspergillus niger, were scanned using a hyperspectral imaging system, and a total of 20 image slices at evenly spaced wavelengths between 1000 to 1600 nm were acquired to form a hypercube. A multivariate image analysis (MIA) technique based on principal component analysis (PCA) was used to reduce the dimensionality of the image hypercubes. Two-class and four-class classification models were developed by applying k-means clustering and discriminant (linear, quadratic, and Mahalanobis) analyses. Two-class discriminant classification models gave maximum classification accuracy of 100%, and on average 97.8% infected kernels were correctly classified by the linear discriminant classifier. The four-class linear discriminant classifier correctly classified more than 95% of the kernels infected with Penicillium and 91.7% healthy kernels. However, the discriminant classifiers misclassified the kernels infected with A. niger and A. glaucus.