Near-infrared hyperspectral imaging and robust statistics for in vivo non-melanoma skin cancer and actinic keratosis characterisation

One of the most common forms of cancer in fair skinned populations is Non-Melanoma Skin Cancer (NMSC), which primarily consists of Basal Cell Carcinoma (BCC), and cutaneous Squamous Cell Carcinoma (SCC). Detecting NMSC early can significantly improve treatment outcomes and reduce medical costs. Similarly, Actinic Keratosis (AK) is a common skin condition that, if left untreated, can develop into more serious conditions, such as SCC. Hyperspectral imagery is at the forefront of research to develop non-invasive techniques for the study and characterisation of skin lesions. This study aims to investigate the potential of near-infrared hyperspectral imagery in the study and identification of BCC, SCC and AK samples in comparison with healthy skin. Here we use a pushbroom hyperspectral camera with a spectral range of ≈ 900 to 1600 nm for the study of these lesions. For this purpose, an ad hoc platform was developed to facilitate image acquisition. This study employed robust statistical methods for the identification of an optimal spectral window where the different samples could be differentiated. To examine these datasets, we first tested for the homogeneity of sample distributions. Depending on these results, either traditional or robust descriptive metrics were used. This was then followed by tests concerning the homoscedasticity, and finally multivariate comparisons of sample variance. The analysis revealed that the spectral regions between 900.66-1085.38 nm, 1109.06-1208.53 nm, 1236.95-1322.21 nm, and 1383.79-1454.83 nm showed the highest differences in this regard, with <1% probability of these observations being a Type I statistical error. Our findings demonstrate that hyperspectral imagery in the near-infrared spectrum is a valuable tool for analyzing, diagnosing, and evaluating non-melanoma skin lesions, contributing significantly to skin cancer research.

Remote Sensing Techniques: Hyperspectral Imaging and Data Analysis

Hyperspectral imaging is a remote sensing technique that enables remote, noninvasive measurement of plant traits. Here, we outline the procedures for camera setup, scanning, and calibration, along with the acquisition of black and white reference materials, which are the key steps in collecting hyperspectral imagery. We also discuss the development of predictive models such as partial least-squares regression, using both large and small datasets, which are used to predict plant traits from hyperspectral data. To ensure practical applicability, we provide code examples that allow readers to immediately implement these techniques in real-world scenarios. We introduce these topics to beginners in an accessible and understandable manner.

Hyperspectral Proximal Sensing for Estimating Photosynthetic Capacities at Leaf and Canopy Scales

Agronomists, plant breeders, and plant biologists have been promoting the need to develop high-throughput methods to measure plant traits of interest for decades. Measuring these plant traits or phenotypes is often a bottleneck since skilled personnel, resources, and ample time are required. Additionally, plant phenotypic traits from only a select number of breeding lines or varieties can be quantified because the "gold standard" measurement of a desired trait cannot be completed in a timely manner. As such, numerous approaches have been developed and implemented to better understand the biology and production of crops and ecosystems. In this chapter, we explain one of the recent approaches leveraging hyperspectral measurements to estimate different aspects of photosynthesis. Notably, we outline the use of hyperspectral radiometer and imaging to rapidly estimate two of the rate-limiting steps of photosynthesis: the maximum rate of the carboxylation of Rubisco (Vcmax) and the maximum rate of electron transfer or regeneration of RuBP (Jmax).

Dynamics and genetic regulation of leaf nutrient concentration in barley based on hyperspectral imaging and machine learning

Biofortification, the enrichment of nutrients in crop plants, is of increasing importance to improve human health. The wild barley nested association mapping (NAM) population HEB-25 was developed to improve agronomic traits including nutrient concentration. Here, we evaluated the potential of high-throughput hyperspectral imaging in HEB-25 to predict leaf concentration of 15 mineral nutrients, sampled from two field experiments and four developmental stages. Particularly accurate predictions were obtained by partial least squares regression (PLS) modeling of leaf concentrations for N, P and K reaching coefficients of determination of 0.90, 0.75 and 0.89, respectively. We recognized nutrient-specific patterns of variation of leaf nutrient concentration between developmental stages. A number of quantitative trait loci (QTL) associated with the simultaneous expression of leaf nutrients were detected, indicating their potential co-regulation in barley. For example, the wild barley allele of QTL-4H-1 simultaneously increased leaf concentration of N, P, K and Cu. Similar effects of the same QTL were previously reported for nutrient concentrations in grains, supporting a potential parallel regulation of N, P, K and Cu in leaves and grains of HEB-25. Our study provides a new approach for nutrient assessment in large-scale field experiments to ultimately select genes and genotypes supporting plant biofortification.

Confocal hyperspectral microscopic imager for the detection and classification of individual microalgae

We propose a confocal hyperspectral microscopic imager (CHMI) that can measure both transmission and fluorescent spectra of individual microalgae, as well as obtain classical transmission images and corresponding fluorescent hyperspectral images with a high signal-to-noise ratio. Thus, the system can realize precise identification, classification, and location of microalgae in a free or symbiosis state. The CHMI works in a staring state, with two imaging modes, a confocal fluorescence hyperspectral imaging (CFHI) mode and a transmission hyperspectral imaging (THI) mode. The imaging modes share the main light path, and thus obtained fluorescence and transmission hyperspectral images have point-to-point correspondence. In the CFHI mode, a confocal technology to eliminate image blurring caused by interference of axial points is included. The CHMI has excellent performance with spectral and spatial resolutions of 3 nm and 2 µm, respectively (using a 10× microscope objective magnification). To demonstrate the capacity and versatility of the CHMI, we report on demonstration experiments on four species of microalgae in free form as well as three species of jellyfish with symbiotic microalgae. In the microalgae species classification experiments, transmission and fluorescence spectra collected by the CHMI were preprocessed using principal component analysis (PCA), and a support vector machine (SVM) model or deep learning was then used for classification. The accuracy of the SVM model and deep learning method to distinguish one species of individual microalgae from another was found to be 96.25% and 98.34%, respectively. Also, the ability of the CHMI to analyze the concentration, species, and distribution differences of symbiotic microalgae in symbionts is furthermore demonstrated.

An evaluation of biochemical, structural and volatile changes of dry-cured pork using a combined ion mobility spectrometry, hyperspectral and confocal imaging approach

BACKGROUND

Food processing induces various modifications that affect the structure, physical and chemical properties of food products and hence the acceptance of the product by the consumer. In this work, the evolution of volatile components, 2-thiobarbituric acid reactive substances (TBARS), moisture content (MC) and microstructural changes of pork was investigated by hyperspectral (HSI) and confocal imaging (CLSM) techniques in synergy with gas chromatography-ion mobility spectrometry (GC-IMS). Models based on partial least squares regression (PLSR) were developed using the full HSI spectrum variables as well as optimum variables selected through a competitive adaptive reweighted sampling algorithm.

RESULTS

Prediction results for MC and TBARS using multiplicative scatter correction pre-processed spectra models demonstrated greater efficiency and predictability with determination coefficient of prediction of 0.928, 0.930 and root mean square error of prediction of 0.114, 1.002, respectively. Major structural changes were also observed during CLSM imaging, which were greatly pronounced in pork samples oven cooked for 15 and 20 h. These structural changes could be related to the denaturation of the major meat components, which could explain the loss of moisture and the formation of TBARS visualized from the HSI chemical distribution maps. GC-IMS identified 35 volatile components, including hexanal and pentanal, which are also known to have a higher lipid oxidation specificity.

CONCLUSION

The synergistic application of HSI, CLSM and GC-IMS enhanced data mining and interpretation and provided a convenient way for analyzing the chemical, structural and volatile changes occurring in meat during processing. © 2021 Society of Chemical Industry.

Nondestructive nitrogen content estimation in tomato plant leaves by Vis-NIR hyperspectral imaging and regression data models

The present study aims to estimate nitrogen (N) content in tomato (Solanum lycopersicum L.) plant leaves using optimal hyperspectral imaging data by means of computational intelligence [artificial neural networks and the differential evolution algorithm (ANN-DE), partial least squares regression (PLSR), and convolutional neural network (CNN) regression] to detect potential plant stress to nutrients at early stages. First, pots containing control and treated tomato plants were prepared; three treatments (categories or classes) consisted in the application of an overdose of 30%, 60%, and 90% nitrogen fertilizer, called N-30%, N-60%, N-90%, respectively. Tomato plant leaves were then randomly picked up before and after the application of nitrogen excess and imaged. Leaf images were captured by a hyperspectral camera, and nitrogen content was measured by laboratory ordinary destructive methods. Two approaches were studied: either using all the spectral data in the visible (Vis) and near infrared (NIR) spectral bands, or selecting only the three most effective wavelengths by an optimization algorithm. Regression coefficients (R) were 0.864±0.027 for ANN-DE, 0.837±0.027 for PLSR, and 0.875±0.026 for CNN in the first approach, over the test set. The second approach used different models for each treatment, achieving R values for all the regression methods above 0.96; however, it needs a previous classification stage of the samples in one of the three nitrogen excess classes under consideration.

Lithological information extraction and classification in hyperspectral remote sensing data using Backpropagation Neural Network

The purposes are to solve the isomorphism encountered while processing hyperspectral remote sensing data and improve the accuracy of hyperspectral remote sensing data in extracting and classifying lithological information. Taking rocks as the research object, Backpropagation Neural Network (BPNN) is introduced. After the hyperspectral image data are normalized, the lithological spectrum and spatial information are the feature extraction targets to construct a deep learning-based lithological information extraction model. The performance of the model is analyzed using specific instance data. Results demonstrate that the overall accuracy and the Kappa coefficient of the lithological information extraction and classification model based on deep learning were 90.58% and 0.8676, respectively. This model can precisely distinguish the properties of rock masses and provide better performance compared with the state of other analysis models. After introducing deep learning, the recognition accuracy and the Kappa coefficient of the proposed BPNN model increased by 8.5% and 0.12, respectively, compared with the traditional BPNN. The proposed extraction and classification model can provide some research values and practical significances for the hyperspectral rock and mineral classification.

Hyperspectral image spectral-spatial classification via weighted Laplacian smoothing constraint-based sparse representation

As a powerful tool in hyperspectral image (HSI) classification, sparse representation has gained much attention in recent years owing to its detailed representation of features. In particular, the results of the joint use of spatial and spectral information has been widely applied to HSI classification. However, dealing with the spatial relationship between pixels is a nontrivial task. This paper proposes a new spatial-spectral combined classification method that considers the boundaries of adjacent features in the HSI. Based on the proposed method, a smoothing-constraint Laplacian vector is constructed, which consists of the interest pixel and its four nearest neighbors through their weighting factor. Then, a novel large-block sparse dictionary is developed for simultaneous orthogonal matching pursuit. Our proposed method can obtain a better accuracy of HSI classification on three real HSI datasets than the existing spectral-spatial HSI classifiers. Finally, the experimental results are presented to verify the effectiveness and superiority of the proposed method.

Hyperspectral imaging in combination with data fusion for rapid evaluation of tilapia fillet freshness

The potential of two different hyperspectral imaging systems (visible near infrared spectroscopy (Vis-NIR) and NIR) was investigated to determine TVB-N contents in tilapia fillets during cold storage. With Vis-NIR and NIR data, calibration models were established between the average spectra of tilapia fillets in the hyperspectral image and their corresponding TVB-N contents and optimized with various variable selection and data fusion methods. Superior models were obtained with variable selection methods based on low-level fusion data when compared with the corresponding methods based on single data blocks. Mid-level fusion data achieved the best model based on CARS, in comparison with all others. Finally, the respective optimized models of single Vis-NIR and NIR data were employed to visualize TVB-N contents distribution in tilapia fillets. In general, the results showed the great feasibility of hyperspectral imaging in combination with data fusion analysis in the nondestructive evaluation of tilapia fillet freshness.

Early prediction of biomass in hybrid rye based on hyperspectral data surpasses genomic predictability in less-related breeding material

Hyperspectral data is a promising complement to genomic data to predict biomass under scenarios of low genetic relatedness. Sufficient environmental connectivity between data used for model training and validation is required. The demand for sustainable sources of biomass is increasing worldwide. The early prediction of biomass via indirect selection of dry matter yield (DMY) based on hyperspectral and/or genomic prediction is crucial to affordably untap the potential of winter rye (Secale cereale L.) as a dual-purpose crop. However, this estimation involves multiple genetic backgrounds and genetic relatedness is a crucial factor in genomic selection (GS). To assess the prospect of prediction using reflectance data as a suitable complement to GS for biomass breeding, the influence of trait heritability ([Formula: see text]) and genetic relatedness were compared. Models were based on genomic (GBLUP) and hyperspectral reflectance-derived (HBLUP) relationship matrices to predict DMY and other biomass-related traits such as dry matter content (DMC) and fresh matter yield (FMY). For this, 270 elite rye lines from nine interconnected bi-parental families were genotyped using a 10 k-SNP array and phenotyped as testcrosses at four locations in two years (eight environments). From 400 discrete narrow bands (410 nm-993 nm) collected by an uncrewed aerial vehicle (UAV) on two dates in each environment, 32 hyperspectral bands previously selected by Lasso were incorporated into a prediction model. HBLUP showed higher prediction abilities (0.41 - 0.61) than GBLUP (0.14 - 0.28) under a decreased genetic relationship, especially for mid-heritable traits (FMY and DMY), suggesting that HBLUP is much less affected by relatedness and [Formula: see text]. However, the predictive power of both models was largely affected by environmental variances. Prediction abilities for DMY were further enhanced (up to 20%) by integrating both matrices and plant height into a bivariate model. Thus, data derived from high-throughput phenotyping emerges as a suitable strategy to efficiently leverage selection gains in biomass rye breeding; however, sufficient environmental connectivity is needed.

Monitoring microstructural changes and moisture distribution of dry-cured pork: a combined confocal laser scanning microscopy and hyperspectral imaging study

BACKGROUND

Various spectral profiles, including reflectance, absorbance, and Kubelka-Munk spectra, have been derived from hyperspectral images and used to develop multivariate models to evaluate changes in the quality of meat and meat products as a function of processing. However, none of these has the capacity to produce images of the structural changes often associated with processing. This study explored the feasibility of combining hyperspectral imaging (HSI) with confocal laser scanning microscopy (CLSM) to examine the impact of processing on microstructural changes and the evolution of moisture. Reflectance spectra features were obtained and transformed into absorbance and Kubelka-Munk spectra and their ability to predict moisture content using models established on partial least-squares regression were evaluated.

RESULTS

The partial least-squares regression model (full-band wavelength) dubbed Rs-MSC yielded the best result, with R c 2 = 0.967 , RMSEC = 0.127, R cv 2 = 0.949 , RMSECV = 0.418, R p 2 = 0.937 , RMSEP = 0.824. Next, a total of 16 optimum wavelengths were selected using the competitive adaptive reweighted sampling algorithm. These wavelengths also yielded good results for Rs-MSC, with R c 2 = 0.958 , RMSEC = 0.840, R cv 2 = 0.931 , RMSECV = 0.118, R p 2 = 0.926 , RMSEP = 0.121. Regarding moisture distribution and microstructure analysis, HSI and CLSM were able to reveal moisture content distribution and conformational differences in microstructure in the test samples.

CONCLUSION

Using HSI in synergy with CLSM may offer a reliable means for assessing both the chemical and structural changes that occur in other congener food products during processing. © 2020 Society of Chemical Industry.

Venom Gland Mass Spectrometry Imaging of Saw-Scaled Viper, Echis carinatus sochureki, at High Lateral Resolution

The snake venom gland is the place for the synthesis, storage, and secretion of a complex mixture of proteins and peptides, i.e., the venom. The morphology of the gland has been revealed by classical histology and microscopic studies. However, knowledge about the gland's cellular secretory and functional processes is still incomplete and has so far been neglected by the omics disciplines. We used autofocusing atmospheric-pressure matrix-assisted laser desorption/ionization (AP-SMALDI) mass spectrometry imaging (MSI) to investigate endogenous biomolecular distributions in the venom glands of the saw-scaled viper, Echis carinatus sochureki, employing different sample preparation methods. Fresh-freezing and formalin-fixation were tested for the gland to obtain intact tissue sections. Subsequently, MSI was conducted with 12 μm pixel resolution for both types of preparations, and the lateral distributions of the metabolites were identified. Experiments revealed that lipids belonging to the classes of PC, SM, PE, PS, PA, and TG are present in the venom gland. PC (32:0) and SM (36:1) were found to be specifically located in the areas where cells are present. The snake venom metalloprotease inhibitor pEKW (m/z 444.2233) was identified in the venom by top-down LC-MS/MS and localized by MALDI-MSI in the gland across secretory epithelial cells. The peptide can inhibit the venom's enzymatic activity during long-term storage within the venom gland. With a high degree of spectral similarities, we concluded that formalin-fixed tissue, in addition to its high ability to preserve tissue morphology, can be considered as an alternative method to fresh-frozen tissue in the case of lipid and peptide MS imaging in venom gland tissues.

Imaging of Neurotransmitters and Small Molecules in Brain Tissues Using Laser Desorption/Ionization Mass Spectrometry Assisted with Zinc Oxide Nanoparticles

Inorganic nanostructured materials such as silicon, carbon, metals, and metal oxides have been explored as matrices of low-background signals to assist the laser desorption/ionization (LDI) mass spectrometric (MS) analysis of small molecules, but their applications for imaging of small molecules in biological tissues remain limited in the literature. Titanium dioxide is one of the known nanoparticles (NP) that can effectively assist LDI MS imaging of low molecular weight molecules (LMWM). TiO₂ NP is commercially available as dispersions, which can be applied using a chemical solution sprayer. However, aggregation of NP can occur in the dispersions, and the aggregated NP can slowly clog the sprayer nozzle. In this work, the use of zinc oxide (ZnO) NP for LDI MS imaging is investigated as a superior alternative due to its dissolution in acidic pH. ZnO NP was found to deliver similar or better results in the imaging of LMWM in comparison to TiO₂ NP. The regular acid washes were effective in minimizing clogging and maintaining high reproducibility. High-quality images of mouse sagittal and rat coronal tissue sections were obtained. Ions were detected predominately as Na⁺ or K⁺ adducts in the positive ion mode. The number of LMWM detected with ZnO NP was similar to that obtained with TiO₂ NP, and only a small degree of specificity was observed.

Evaluation of Dianhong black tea quality using near-infrared hyperspectral imaging technology

BACKGROUND

Tea (Camellia sinensis L) is a highly nutritious beverage with commercial value globally. However, it is at risk of economic fraud. This study aims to develop a powerful evaluation method to distinguish Chinese official Dianhong tea from various other categories, employing hyperspectral imaging (HSI) technology and chemometric algorithms.

RESULTS

Two matrix statistical algorithms encompassing a gray-level co-occurrence matrix (GLCM) and a gradient co-occurrence matrix (GLGCM) are used to extract HSI texture data. Three novel spectral variable screening methods are utilized to select wavenumbers of near-infrared (NIR) spectra: iteratively retaining informative variables (IRIV), interval random frog, and variable combination population analysis. Feature fusion of image texture characteristics and spectra data are the eigenvectors for model building. Authentic classification models are constructed using the extreme learning machine approach and the least squares support vector machine (LSSVM) approach, coupling them with features from wavelength extraction techniques for assessing the quality of Dianhong black tea. The results demonstrate that the LSSVM model using fused data (IRIV + GLGCM) provides the best results and achieves a predictive precision of 99.57%.

CONCLUSION

This study confirms that HSI coupled with LSSVM is effective in differentiating authentic Dianhong black tea samples. © 2020 Society of Chemical Industry.

Anisotropic effect on the predictability of intramuscular fat content in pork by hyperspectral imaging and chemometrics

The fibrous structure of meat muscle makes it an anisotropic optical material. As such, spectral information varies with the orientation of the muscle. In this study, spectral data from pork cuts were obtained by a transverse scan (TRANSCAN), radial scan (RADISCAN), and longitudinal scan (LONGSCAN) by using hyperspectral imaging. The information was used to develop and compare the prediction models for intramuscular (IMF) content prediction by partial least square regression (PLSR), support vector machines regression (SVMR), and backpropagation artificial neural network (BPANN). The three modeling algorithms showed equal capability for modeling IMF in pork. The accuracy of the prediction models from the three scans was in the order of TRANSCAN ≥ RADISCAN ≥ LONGSCAN. Successive projection algorithm reduced the wavelengths to 93%. The reduced wavelengths were used to build new models that showed similar accuracy to the models of the original wavelengths. This study shows that muscle orientation influences the accuracy of the prediction models.

Prediction of dry matter content of recently harvested 'Hass' avocado fruits using hyperspectral imaging

BACKGROUND

'Hass' avocado consumption is increasing due to its organoleptic properties, so it is necessary to develop new technologies to guarantee export quality. Avocado fruits do not ripen on the tree, and the visual classification of its maturity is not accurate. The most commonly used fruit maturity indicator is the percentage of dry matter (DM). The aim of this research was to investigate a non-destructive method with hyperspectral images to predict the percentage of DM of fruits across the spectral range of 400-1000 nm.

RESULTS

No correlation between fruit weight and color with the percentage of DM was found in the study area. Cross-validation efficiency of different data sources, including the spectrum extraction zone (the center, a line from the peduncle to the base, and the whole fruit) and the average of one or two fruit faces, was compared. Four linear regression models were compared. Data of the whole fruit and average of both sides per fruit using a support vector machine regression were selected for the prediction test. Following the cross-validation concept, five sets of calibration and test data were selected and optimized for calibration. The best test prediction set comprised an R² = 0.9, a root-mean-square error of 2.6 g kg^-1 DM, a Pearson correlation of 0.95, and a ratio of prediction to deviation of 3.2.

CONCLUSIONS

The results of the study indicate that hyperspectral images allow classifying export fruits and making harvesting decisions. © 2020 Society of Chemical Industry.

Identification of plant leaf phosphorus content at different growth stages based on hyperspectral reflectance

BACKGROUND

Modern agriculture strives to sustainably manage fertilizer for both economic and environmental reasons. The monitoring of any nutritional (phosphorus, nitrogen, potassium) deficiency in growing plants is a challenge for precision farming technology. A study was carried out on three species of popular crops, celery (Apium graveolens L., cv. Neon), sugar beet (Beta vulgaris L., cv. Tapir) and strawberry (Fragaria × ananassa Duchesne, cv. Honeoye), fertilized with four different doses of phosphorus (P) to deliver data for non-invasive detection of P content.

RESULTS

Data obtained via biochemical analysis of the chlorophyll and carotenoid contents in plant material showed that the strongest effect of P availability for plants was in the diverse total chlorophyll content in sugar beet and celery compared to that in strawberry, in which P affects a variety of carotenoid contents in leaves. The measurements performed using hyperspectral imaging, obtained in several different stages of plant development, were applied in a supervised classification experiment. A machine learning algorithm (Backpropagation Neural Network, Random Forest, Naive Bayes and Support Vector Machine) was developed to classify plants from four variants of P fertilization. The lowest prediction accuracy was obtained for the earliest measured stage of plant development. Statistical analyses showed correlations between leaf biochemical constituents, phosphorus fertilization and the mass of the leaf/roots of the plants.

CONCLUSIONS

Obtained results demonstrate that hyperspectral imaging combined with artificial intelligence methods has potential for non-invasive detection of non-homogenous phosphorus fertilization on crop levels.

Bread and durum wheat classification using wavelet based image fusion

BACKGROUND

Wheat, which is an essential nutrient, is an important food source for human beings because it is used in flour and feed production. As in many nutrients, wheat plays an important role in macaroni and bread production. The types of wheat used for both foods are different, namely bread and durum wheat. A strong separation of these two wheat types is important for product quality. This article differs from the traditional methods available for the identification of bread and durum wheat species. In this study, ultraviolet (UV) and white light (WL) images of wheat are obtained for both species. Wheat types in these images are classified by various machine learning (ML) methods. Afterwards, these images are fused by wavelet-based image fusion method.

RESULTS

The highest accuracy value calculated using only UV and only WL image is 94.8276% and these accuracies are obtained by Support Vector Machine (SVM) and multilayer perceptron (MLP) algorithms, respectively. However, this accuracy value is 98.2759% for the fusion image and both MLP and SVM achieved the same success.

CONCLUSION

Wavelet-based fusion has increased the classification accuracy of all three learning algorithms. It is concluded that the identification ability in the resulting fusion image is higher than the other two raw images. © 2020 Society of Chemical Industry.

Using near-infrared hyperspectral imaging with multiple decision tree methods to delineate black tea quality

The evaluation of tea quality tended to be subjective and empirical by human panel tests currently. A convenient analytical approach without human involvement was developed for the quality assessment of tea with great significance. In this study, near-infrared hyperspectral imaging (HSI) combined with multiple decision tree methods was utilized as an objective analysis tool for delineating black tea quality and rank. Data fusion that integrated texture features based on gray-level co-occurrence matrix (GLCM) and short-wave near-infrared spectral features were as the target characteristic information for modeling. Three different types of supervised decision tree algorithms (fine tree, medium tree, and coarse tree) were proposed for the comparison of the modeling effect. The results indicated that the performance of models was enhanced by the multiple perception feature fusion. The fine tree model based on data fusion obtained the best predictive performance, and the correct classification rate (CCR) of evaluating black tea quality was 93.13% in the prediction process. This work demonstrated that HSI coupled with intelligence algorithms as a rapid and effective strategy could be successfully applied to accurately identify the rank quality of black tea.

Detection and identification of Cannabis sativa L. using near infrared hyperspectral imaging and machine learning methods. A feasibility study

Remote identification of illegal plantations of Cannabis sativa Linnaeus is an important task for the Brazilian Federal Police. The current analytical methodology is expensive and strongly dependent on the expertise of the forensic investigator. A faster and cheaper methodology based on automatic methods can be useful for the detection and identification of Cannabis sativa L. in a reliable and objective manner. In this work, the high potential of Near Infrared Hyperspectral Imaging (HSI-NIR) combined with machine learning is demonstrated for supervised detection and classification of Cannabis sativa L. This plant, together with other plants commonly found in the surroundings of illegal plantations and soil, were directly collected from an illegal plantation. Due to the high correlation of the NIR spectra, sparse Principal Component Analysis (sPCA) was implemented to select the most important wavelengths for identifying Cannabis sativa L. One class Soft Independent Class Analogy model (SIMCA) was built, considering just the spectral variables selected by sPCA. Sensitivity and specificity values of 89.45% and 97.60% were, respectively, obtained for an external validation set subjected to the s-SIMCA. The results proved the reliability of a methodology based on NIR hyperspectral cameras to detect and identify Cannabis sativa L., with only four spectral bands, showing the potential of this methodology to be implemented in low-cost airborne devices.

Integration of spectroscopy and image for identifying fusarium damage in wheat kernels

Hyperspectral imaging (HSI) was studied for the detection of varying degrees of damage in wheat kernels caused by Fusarium head blight (Gibberella zeae), a major disease in wheat worldwide. A total of 810 wheat kernel samples were collected from a field trial with the three levels of Fusarium infection, healthy, moderate, and severe. Hyperspectral image of the wheat kernels was acquired over a wavelength range of 400-1000 nm. The raw spectral data were pre-processed, and then the optimal wavelengths were selected using principal component analysis (PCA), successive projection algorithm (SPA) and random forest (RF). The image features were extracted based on the optimal wavelengths, and then the spectral features and image features were combined as fusion features. Support vector machine (SVM), random forest (RF) and naive Bayes (NB) were employed to build the classification models to identify the degrees of Fuasrium damage based on spectral and fusion features. The best performance was obtained by using the SPA-RF method to select the optimal wavelengths and corresponding image features, with a classification accuracy of 96.44%. The method developed from this study can provide a more effective way to identify the degrees of Fusarium damage in wheat kernels.

Rapid detection of quality index of postharvest fresh tea leaves using hyperspectral imaging

BACKGROUND

The quality of fresh tea leaves after harvest determines, to some extent, the quality and price of commercial tea. A fast and accurate method to evaluate the quality of fresh tea leaves is required.

RESULTS

In this study, the potential of hyperspectral imaging in the range of 328-1115 nm for the rapid prediction of moisture, total nitrogen, crude fiber contents, and quality index value was investigated. Ninety samples of eight tea-leaf varieties and two picking standards were tested. Quantitative partial least squares regression (PLSR) models were established using a full spectrum, whereas multiple linear regression (MLR) models were developed using characteristic wavelengths selected by a successive projections algorithm (SPA) and competitive adaptive reweighted sampling. The results showed that the optimal SPA-MLR models for moisture, total nitrogen, crude fiber contents, and quality index value yielded optimal performance with coefficients of determination for prediction (R² p) of 0.9357, 0.8543, 0.8188, 0.9168; root mean square error of 0.3437, 0.1097, 0.3795, 1.0358; and residual prediction deviation of 4.00, 2.56, 2.31, and 3.51, respectively.

CONCLUSION

The results suggested that the hyperspectral imaging technique coupled with chemometrics was a promising tool for the rapid and nondestructive measurement of tea-leaf quality, and had the potential to develop multispectral imaging systems for future online detection of tea-leaf quality. © 2020 Society of Chemical Industry.

Combination of spectra and texture data of hyperspectral imaging for prediction and visualization of palmitic acid and oleic acid contents in lamb meat

The feasibility of combining spectral and textural information from hyperspectral imaging to improve the prediction of the C16:0 and C18:1 n9 contents for lamb was explored. 29 and 22 optimal wavelengths were selected for the C16:0 and C18:1 n9 contents, respectively, by conducting the variable combination population analysis-iteratively retaining informative variables (VCPA-IRIV) algorithm. To extract the textural features of images, a gray-level co-occurrence matrix (GLCM) analysis was implemented in the first principal component image. The least squares support vector machine (LSSVM) model and the partial least squares regression (PLSR) model were developed to predict the C16:0 and C18:1 n9 contents from the spectra and the fusion data. The distribution map was visualized using the best model with the imaging process. The results showed that the combination of the spectral and textural information of hyperspectral imaging coupled with the VCPA-IRIV algorithm had strong potential for the prediction and visualization of the C16:0 and C18:1 n9 contents of lamb.

Estimating photosynthetic traits from reflectance spectra: A synthesis of spectral indices, numerical inversion, and partial least square regression

The lack of efficient means to accurately infer photosynthetic traits constrains understanding global land carbon fluxes and improving photosynthetic pathways to increase crop yield. Here, we investigated whether a hyperspectral imaging camera mounted on a mobile platform could provide the capability to help resolve these challenges, focusing on three main approaches, that is, reflectance spectra-, spectral indices-, and numerical model inversions-based partial least square regression (PLSR) to estimate photosynthetic traits from canopy hyperspectral reflectance for 11 tobacco cultivars. Results showed that PLSR with inputs of reflectance spectra or spectral indices yielded an R2 of ~0.8 for predicting V cmax and J max , higher than an R2 of ~0.6 provided by PLSR of numerical inversions. Compared with PLSR of reflectance spectra, PLSR with spectral indices exhibited a better performance for predicting V cmax (R2 = 0.84 ± 0.02, RMSE = 33.8 ± 2.2 μmol m-2  s-1 ) while a similar performance for J max (R2 = 0.80 ± 0.03, RMSE = 22.6 ± 1.6 μmol m-2  s-1 ). Further analysis on spectral resampling revealed that V cmax and J max could be predicted with ~10 spectral bands at a spectral resolution of less than 14.7 nm. These results have important implications for improving photosynthetic pathways and mapping of photosynthesis across scales.

Development and Evaluation of a New Spectral Disease Index to Detect Wheat Fusarium Head Blight Using Hyperspectral Imaging

Fusarium head blight (FHB) is a major disease threatening worldwide wheat production. FHB is a short cycle disease and is highly destructive under conducive environments. To provide technical support for the rapid detection of the FHB disease, we proposed to develop a new Fusarium disease index (FDI) based on the spectral data of 374–1050 nm. This study was conducted through the analysis of reflectance spectral data of healthy and diseased wheat ears at the flowering and filling stages by hyperspectral imaging technology and the random forest method. The characteristic wavelengths selected were 570 nm and 678 nm for the late flowering stage, 565 nm and 661 nm for the early filling stage, 560 nm and 663 nm for the combined stage (combining both flowering and filling stages) by random forest. FDI at each stage was derived from the wavebands of each corresponding stage. Compared with other 16 existing spectral indices, FDI demonstrated a stronger ability to determine the severity of the FHB disease. Its determination coefficients (R²) values exceeded 0.90 and the RMSEs were less than 0.08 in the models for each stage. Furthermore, the model for the combined stage performed better when used at single growth stage, but its effect was weaker than that of the models for the two individual growth stages. Therefore, using FDI can provide a new tool to detect the FHB disease at different growth stages in wheat.

Unified Classification of Bacterial Colonies on Different Agar Media Based on Hyperspectral Imaging and Machine Learning

A universal method by considering different types of culture media can enable convenient classification of bacterial species. The study combined hyperspectral technology and versatile chemometric algorithms to achieve the rapid and non-destructive classification of three kinds of bacterial colonies (Escherichia coli, Staphylococcus aureus and Salmonella) cultured on three kinds of agar media (Luria–Bertani agar (LA), plate count agar (PA) and tryptone soy agar (TSA)). Based on the extracted spectral data, partial least squares discriminant analysis (PLS-DA) and support vector machine (SVM) were employed to established classification models. The parameters of SVM models were optimized by comparing genetic algorithm (GA), particle swarm optimization (PSO) and grasshopper optimization algorithm (GOA). The best classification model was GOA-SVM, where the overall correct classification rates (OCCRs) for calibration and prediction of the full-wavelength GOA-SVM model were 99.45% and 98.82%, respectively, and the Kappa coefficient for prediction was 0.98. For further investigation, the CARS, SPA and GA wavelength selection methods were used to establish GOA-SVM simplified model, where CARS-GOA-SVM was optimal in model accuracy and stability with the corresponding OCCRs for calibration and prediction and the Kappa coefficients of 99.45%, 98.73% and 0.98, respectively. The above results demonstrated that it was feasible to classify bacterial colonies on different agar media and the unified model provided a continent and accurate way for bacterial classification.

Hyperspectral imaging combined with machine learning as a tool to obtain high-throughput plant salt-stress phenotyping

The rapid selection of salinity-tolerant crops to increase food production in salinized lands is important for sustainable agriculture. Recently, high-throughput plant phenotyping technologies have been adopted that use plant morphological and physiological measurements in a non-destructive manner to accelerate plant breeding processes. Here, a hyperspectral imaging (HSI) technique was implemented to monitor the plant phenotypes of 13 okra (Abelmoschus esculentus L.) genotypes after 2 and 7 days of salt treatment. Physiological and biochemical traits, such as fresh weight, SPAD, elemental contents and photosynthesis-related parameters, which require laborious, time-consuming measurements, were also investigated. Traditional laboratory-based methods indicated the diverse performance levels of different okra genotypes in response to salinity stress. We introduced improved plant and leaf segmentation approaches to RGB images extracted from HSI imaging based on deep learning. The state-of-the-art performance of the deep-learning approach for segmentation resulted in an intersection over union score of 0.94 for plant segmentation and a symmetric best dice score of 85.4 for leaf segmentation. Moreover, deleterious effects of salinity affected the physiological and biochemical processes of okra, which resulted in substantial changes in the spectral information. Four sample predictions were constructed based on the spectral data, with correlation coefficients of 0.835, 0.704, 0.609 and 0.588 for SPAD, sodium concentration, photosynthetic rate and transpiration rate, respectively. The results confirmed the usefulness of high-throughput phenotyping for studying plant salinity stress using a combination of HSI and deep-learning approaches.

Low-cost hyper-spectral imaging system using a linear variable bandpass filter for agritech applications

Hyperspectral imaging for agricultural applications provides a solution for non-destructive, large-area crop monitoring. However, current products are bulky and expensive due to complicated optics and electronics. A linear variable filter was developed for implementation into a prototype hyperspectral imaging camera that demonstrates good spectral performance between 450 and 900 nm. Equipped with a feature extraction and classification algorithm, the proposed system can be used to determine potato plant health with ∼88% accuracy. This algorithm was also capable of species identification and is demonstrated as being capable of differentiating between rocket, lettuce, and spinach. Results are promising for an entry-level, low-cost hyperspectral imaging solution for agriculture applications.

Visible near infrared reflectance molecular chemical imaging of human ex vivo carcinomas and murine in vivo carcinomas

SIGNIFICANCE

A key risk faced by oncological surgeons continues to be complete removal of tumor. Currently, there is no intraoperative imaging device to detect kidney tumors during excision.

AIM

We are evaluating molecular chemical imaging (MCI) as a technology for real-time tumor detection and margin assessment during tumor removal surgeries.

APPROACH

In exploratory studies, we evaluate visible near infrared (Vis-NIR) MCI for differentiating tumor from adjacent tissue in ex vivo human kidney specimens, and in anaesthetized mice with breast or lung tumor xenografts. Differentiation of tumor from nontumor tissues is made possible with diffuse reflectance spectroscopic signatures and hyperspectral imaging technology. Tumor detection is achieved by score image generation to localize the tumor, followed by application of computer vision algorithms to define tumor border.

RESULTS

Performance of a partial least squares discriminant analysis (PLS-DA) model for kidney tumor in a 22-patient study is 0.96 for area under the receiver operating characteristic curve. A PLS-DA model for in vivo breast and lung tumor xenografts performs with 100% sensitivity, 83% specificity, and 89% accuracy.

CONCLUSION

Detection of cancer in surgically resected human kidney tissues is demonstrated ex vivo with Vis-NIR MCI, and in vivo on mice with breast or lung xenografts.

HYPerspectral Enhanced Reality (HYPER): a physiology-based surgical guidance tool

BACKGROUND

HSI is an optical technology allowing for a real-time, contrast-free snapshot of physiological tissue properties, including oxygenation. Hyperspectral imaging (HSI) has the potential to quantify the gastrointestinal perfusion intraoperatively. This experimental study evaluates the accuracy of HSI, in order to quantify bowel perfusion, and to obtain a superposition of the hyperspectral information onto real-time images.

METHODS

In 6 pigs, 4 ischemic bowel loops were created (A, B, C, D) and imaged at set time points (from 5 to 360 min). A commercially available HSI system provided pseudo-color maps of the perfusion status (StO2, Near-InfraRed perfusion) and the tissue water index. An ad hoc software was developed to superimpose HSI information onto the live video, creating the HYPerspectral-based Enhanced Reality (HYPER). Seven regions of interest (ROIs) were identified in each bowel loop according to StO2 ranges, i.e., vascular (VASC proximal and distal), marginal vascular (MV proximal and distal), marginal ischemic (MI proximal and distal), and ischemic (ISCH). Local capillary lactates (LCL), reactive oxygen species (ROS), and histopathology were measured at the ROIs. A machine-learning-based prediction algorithm of LCL, based on the HSI-StO2%, was trained in the 6 pigs and tested on 5 additional animals.

RESULTS

HSI parameters (StO2 and NIR) were congruent with LCL levels, ROS production, and histopathology damage scores at the ROIs discriminated by HYPER. The global mean error of LCL prediction was 1.18 ± 1.35 mmol/L. For StO2 values > 30%, the mean error was 0.3 ± 0.33.

CONCLUSIONS

HYPER imaging could precisely quantify the overtime perfusion changes in this bowel ischemia model.