Hyperspectral Research Articles

Chestnut quality classification by THz Time-Domain Hyperspectral Imaging combined with unsupervised learning analysis

Chestnut crops are threatened by fungal pathogens such as Gnomoniopsis castaneae, which cause significant degradation of quality. Early detection of such infections is crucial to maintain the quality of chestnuts in the food industry. This study explores the application of Terahertz Time-Domain Hyperspectral Imaging (THz-TDHIS) combined with unsupervised learning techniques to identify fungal infections in chestnuts. Unlike conventional methods that rely on light attenuation, this approach leverages the unique spectral signatures of infected tissues. By employing Principal Component Analysis, K-Means Clustering, and Agglomerative Clustering, we effectively differentiate between healthy and infected portions of chestnuts. Our findings indicate that spectral features, rather than just intensity variations, provide more reliable markers for infection. In addition, we demonstrate that these methods enable the quantification of the degree of infection in chestnuts. The robustness of these unsupervised learning methods in handling large and heterogeneous data sets further underscores their potential in agricultural applications. This integrated THz-TDHIS and machine learning approach presents a promising solution to ensure chestnut quality and safety.

Application of Supervised Learning Methods and Information Gain Methods in the Determination of Asbestos–Cement Roofs’ Deterioration State

This study introduces an innovative approach to evaluate the condition of asbestos–cement (AC) roofs by integrating field data with five distinct supervised learning models: decision trees, KNN, logistic regression, support vector machine, and random forest. A novel methodology for assessing the importance of 380 reflectance bands was employed, offering fresh insights into the key indicators of AC roof deterioration. The research systematically organized and prioritized reflectance bands based on their information gain, optimizing both the selection of relevant bands and the performance of the models in differentiating between low and high intervention priority (LIP and HIP) roofs. The decision tree model, when applied to the top 10 most relevant bands, achieved the highest cross-validation accuracy of 76.047%, making it the most effective tool for identifying AC roof conditions. Additionally, the random forest model demonstrated strong performance across various band groups, further validating its utility. Utilizing the open-source software Weka (version 3.8.6), this study adeptly executed relevance evaluation and model implementation, providing a practical and scalable solution for material characterization, especially in regions where resources for spectral and hyperspectral image analysis are limited. The findings of this study offer valuable tools for government and environmental authorities, particularly in developing countries, where efficient and cost-effective AC roof assessment is crucial for public health and safety. The methodology is adaptable to different urban environments and climatic conditions, supporting global efforts in asbestos management, especially in countries where asbestos regulations are newly implemented. Organized within the CRISP-DM framework, this paper details the methodological phases, presents compelling results on reflectance band relevance, evaluates machine learning models, and concludes with prospects for future research aimed at enhancing asbestos detection and removal strategies.

Spectral Analysis on Color Detection Sharpness of Animal Vision toward Polychromatic Vision System

AbstractSpectral discrimination by animal visions such as human, bird, and butterfly is numerically compared by angular analysis of photo‐response (PR) which will be recorded at photo‐receptors. Hyperspectral imaging system is utilized to simulate various animal vision. Bird vision is acute to discriminate colors among different vegetables due to its evenly spaced and narrow spectral responsivity of photo‐receptors compared to that of human. Butterfly vision is excellent in discriminating red tomato ripening due to the exclusive photo‐receptors detecting only over 600 nm. Even real and fake fruits in the same perceived color for human is discriminated by bird vision. Artificial vision with finely resolved polychromatic artificial cone cells are demonstrated surpassing human vision using visible multispectral camera. This provides insights on designing novel bio‐inspired vision system.

EELS hyperspectral images unmixing - using autoencoders

Spatially resolved Electron Energy-Loss Spectroscopy conducted in a Scanning Transmission Electron Microscope enables the acquisition of hyperspectral images. Spectral unmixing is the process of decomposing each spectrum of an hyperspectral image into a combination of representative spectra (endmembers) corresponding to compounds present in the sample along with their local proportions (abundances). Spectral unmixing is a complex task, and various methods have been developed in different communities using hyperspectral images. However, none of these methods fully satisfy the spatially resolved Electron Energy-Loss Spectroscopy requirements. Recent advancements in remote sensing, which focus on Deep Learning techniques, have the potential to meet these requirements, particularly Autoencoders. In this study, the performance of Deep Learning methods using autoencoders for spectral unmixing is evaluated, and their results are compared with traditional methods. Synthetic hyperspectral images have been created to quantitatively assess the outcomes of the unmixing process using specific metrics. The methods are subsequently applied to a series of experimental data. The findings demonstrate the promising potential of autoencoders as a tool for Electron Energy-Loss Spectroscopy hyperspectral images unmixing, marking a starting point for exploring more sophisticated Neural Networks.

Evaluation of a hyperspectral image pipeline toward building a generalisation capable crop dry matter content prediction model

Hyperspectral imaging has proven to be a reliable technique for estimating dry matter, a common variable when considering the quality of the fresh produce. However, developing models capable of generalising across different crops is challenging. In this study, several pipelines were explored towards achieving a robust and accurate generic regression model were evaluated and the development of Automatic Relevance Determination (ARD) and Partial Least Squares (PLS) algorithms for fruit and vegetable dry matter estimation. The models were built using a VIS-NIR dataset that includes both fruit and vegetables, namely, apples, broccoli and leek (n = 779). The PLS regression model obtained Root Mean Square on Prediction (RMSEP) = 0.0137, outperforming ARD regression (RMSEP = 0.0140) on a 10x5-fold cross-validation protocol. The evaluated preprocessing techniques affect the two regression algorithms differently, with the best results achieved when the pipeline was used without feature extraction. Overall, the pipeline using either ARD or PLS regression shows strong performance and generalisation for Visible-Near Infrared (VIS-NIR)-based dry matter estimation across diverse fruits and vegetables.

A novel low-quality Raman hyperspectral image reconstruction method for corn component mapping

Raman imaging used to obtain spatial distribution information of chemical components on the sample surface has been widely applied in the nutritional assessment and quality control of cereal grains. However, obtaining high-quality Raman hyperspectral images of cereal grain based on point scanning is time-consuming due to the limitations of the instrumental imaging principle and Raman hyperspectral images acquired in a short time are low-quality with noise interference, limiting its practical application in the visualization of cereal grain composition. To rapidly and effectively achieve high-quality component distribution mappings of cereal grains based on Raman hyperspectral images, we propose a novel Swin Transformer-based method combining spatial-spectral denoising and image reconstruction (SwinSSID-IR) for low-quality Raman hyperspectral image reconstruction. The novel denoising network named SwinSSID innovatively introduces Swin Transformer blocks (STB) to effectively eliminate noise from raw low-quality Raman hyperspectral images of corn kernels. Next, the super-resolution network named SwinIR rapidly generates high-quality component distribution mappings of corn with exceptional clarity and detail from the denoised low-quality Raman hyperspectral image at characteristic peak. Experimental results show the competitiveness of the proposed method compared with other similar advanced methods and its effective application in non-destructive visual analysis of corn components.

Monitoring of plant diseases caused by Fusarium commune and Rhizoctonia solani in bok choy using hyperspectral remote sensing and machine learning.

Local vegetable production is susceptible to various fungal pathogens, the most common and lethal of which are Fusarium commune and Rhizoctonia solani. Early detection of these pathogens is challenging, and by the time visual symptoms appear, the pathogens may have already spread extensively, causing massive damage to the production. In this study, we explored the use of hyperspectral data for early detection of diseases caused by F. commune or R. solani in bok choy Brassica rapa subsp. chinensis by collecting hyperspectral data from healthy plants and plants inoculated with either fungal pathogen in a controlled experimental set-up. Based on the collected data, we employed various tree-based, distribution-based, geometric, neural networks and ensemble learning algorithms to train detection models. Among the trained models, Multi-Layer Perceptron (MLP) models performed the best with overall accuracy reaching 95.9 ± 0.26%. MLP models could differentiate between healthy and infected plants with 99% precision after 1 day of infection, and distinguish between different fungal pathogens with 99% precision after 2 days. During this period, no visible symptoms of fungal infection could be observed. Further analysis into trained MLP models and general reflectance profiles of plants also revealed a high correlation of the spectral regions 445-460, 560-595, 606-620 and 719-728 nm with fungal infection in bok choy plants. Our findings highlight the potential of hyperspectral imaging as a highly precise early detection tool for fungal diseases in plants. © 2024 Society of Chemical Industry.

SMALE: Hyperspectral Image Classification via Superpixels and Manifold Learning

As an extremely efficient preprocessing tool, superpixels have become more and more popular in various computer vision tasks. Nevertheless, there are still several drawbacks in the application of hyperspectral image (HSl) processing. Firstly, it is difficult to directly apply superpixels because of the high dimension of HSl information. Secondly, existing superpixel algorithms cannot accurately classify the HSl objects due to multi-scale feature categorization. For the processing of high-dimensional problems, we use the principle of PCA to extract three principal components from numerous bands to form three-channel images. In this paper, a novel superpixel algorithm called Seed Extend by Entropy Density (SEED) is proposed to alleviate the seed point redundancy caused by the diversified content of HSl. It also focuses on breaking the dilemma of manually setting the number of superpixels to overcome the difficulty of classification imprecision caused by multi-scale targets. Next, a space–spectrum constraint model, termed Hyperspectral Image Classification via superpixels and manifold learning (SMALE), is designed, which integrates the proposed SEED to generate a dimensionality reduction framework. By making full use of spatial context information in the process of unsupervised dimension reduction, it could effectively improve the performance of HSl classification. Experimental results show that the proposed SEED could effectively promote the classification accuracy of HSI. Meanwhile, the integrated SMALE model outperforms existing algorithms on public datasets in terms of several quantitative metrics.

Hyperspectral Imaging for Phenotyping Plant Drought Stress and Nitrogen Interactions Using Multivariate Modeling and Machine Learning Techniques in Wheat

Accurate detection of drought stress in plants is essential for water use efficiency and agricultural output. Hyperspectral imaging (HSI) provides a non-invasive method in plant phenotyping, allowing the long-term monitoring of plant health due to sensitivity to subtle changes in leaf constituents. The broad spectral range of HSI enables the development of different vegetation indices (VIs) to analyze plant trait responses to multiple stresses, such as the combination of nutrient and drought stresses. However, known VIs may underperform when subjected to multiple stresses. This study presents new VIs in tandem with machine learning models to identify drought stress in wheat plants under varying nitrogen (N) levels. A pot wheat experiment was set up in the glasshouse with four treatments: well-watered high-N (WWHN), well-watered low-N (WWLN), drought-stress high-N (DSHN) and drought-stress low-N (DSLN). In addition to ensuring that plants were watered according to the experiment design, photosynthetic rate (Pn) and stomatal conductance (gs) (which are used to assess plant drought stress) were taken regularly, serving as the ground truth data for this study. The proposed VIs, together with known VIs, were used to train three classification models: support vector machines (SVM), random forest (RF), and deep neural networks (DNN) to classify plants based on their drought status. The proposed VIs achieved more than 0.94 accuracy across all models, and their performance further increased when combined with known VIs. The combined VIs were used to train three regression models to predict the stomatal conductance and photosynthetic rates of plants. The random forest regression model performed best, suggesting that it could be used as a stand-alone tool to forecast gs and Pn and track drought stress in wheat. This study shows that combining hyperspectral data with machine learning can effectively monitor and predict drought stress in crops, especially in varying nitrogen conditions.

Low computational demand stray light correction method for hyperspectral imaging spectrometers

Stray light correction in hyperspectral imaging spectrometers has long been restricted by high computational requirements. This paper presents a low computational demand method, based on the matrix operations for spectrometer stray light correction, using an iterative approach to efficiently correct stray light across both spectral and spatial dimensions. The efficacy of this method is demonstrated through its application to simulated and real images, achieving an overall reduction of stray light by over 50%, with significantly reduced computation time and memory usage compared to the method by Zong et al. based on a sparse matrix [Appl. Opt. 45, 1111 (2006)10.1364/AO.45.001111APOPAI2155-3165]. By enabling stray light correction on general-purpose computers, this method enhances affordability and accessibility, promoting broader use and reducing measurement uncertainties in various hyperspectral imaging applications.

Machine Learning for Deconvolution and Segmentation of Hyperspectral Imaging Data from Biopharmaceutical Resins.

Biopharmaceutical resins are pivotal inert matrices used across industry and academia, playing crucial roles in a myriad of applications. For biopharmaceutical process research and development applications, a deep understanding of the physical and chemical properties of the resin itself is frequently required, including for drug purification, drug delivery, and immobilized biocatalysis. Nevertheless, the prevailing methodologies currently employed for elucidating these important aspects of biopharmaceutical resins are often lacking, frequently require significant sample alteration, are destructive or ionizing in nature, and may not adequately provide representative information. In this work, we propose the use of unsupervised machine learning technologies, in the form of both non-negative matrix factorization (NMF) and k-means segmentation, in conjugation with Raman hyperspectral imaging to rapidly elucidate the molecular and spatial properties of biopharmaceutical resins. Leveraging our proposed technology, we offer a new approach to comprehensively understanding important resin-based systems for application across biopharmaceuticals and beyond. Specifically, focusing herein on a representative resin widely utilized across the industry (i.e., Immobead 150P), our findings showcase the ability of our machine learning-based technology to molecularly identify and spatially resolve all chemical species present. Further, we offer a comprehensive evaluation of optimal excitation for hyperspectral imaging data collection, demonstrating results across 532, 638, and 785 nm excitation. In all cases, our proposed technology deconvoluted, both spatially and spectrally, resin and glass substrates via NMF. After NMF deconvolution, image segmentation was also successfully accomplished in all data sets via k-means clustering. To the best of our knowledge, this is the first report utilizing the combination of two unsupervised machine learning methodologies, combining NMF and k-means, for the rapid deconvolution and segmentation of biopharmaceutical resins. As such, we offer a powerful new data-rich experimentation tool for application across multidisciplinary fields for a deeper understanding of resins.

Blind non-linear spectral unmixing with spatial coherence for hyper and multispectral images

Multi and hyperspectral images have become invaluable sources of information, revolutionizing various fields such as remote sensing, environmental monitoring, agriculture and medicine. In this expansive domain, the multi-linear mixing model (MMM) is a versatile tool to analyze spatial and spectral domains by effectively bridging the gap between linear and non-linear interactions of light and matter. This paper introduces an upgraded methodology that integrates the versatility of MMM in non-linear spectral unmixing, while leveraging spatial coherence (SC) enhancement through total variation theory to mitigate noise effects in the abundance maps. Referred to as non-linear extended blind end-member and abundance extraction with SC (NEBEAE-SC), the proposed methodology relies on constrained quadratic optimization, cyclic coordinate descent algorithm, and the split Bregman formulation. The validation of NEBEAE-SC involved rigorous testing on various hyperspectral datasets, including a synthetic image, remote sensing scenarios, and two biomedical applications. Specifically, our biomedical applications are focused on classification tasks, the first addressing hyperspectral images of in-vivo brain tissue, and the second involving multispectral images of ex-vivo human placenta. Our results demonstrate an improvement in the abundance estimation by NEBEAE-SC compared to similar algorithms in the state-of-the-art by offering a robust tool for non-linear spectral unmixing in diverse application domains.

Detection of physical hazards from fruit processed products using hyperspectral imaging and prediction based on PLS-DA and logistic regression machine learning models

The current study aims to investigate a prediction model from reflection values and the spectral angle mapper (SAM) obtained from hyperspectral imaging (HSI) technology for the detection of three types of foreign materials – a branch, a knife, and rubber – that are associated with the various fruit-processed products. The study found that the maximum and minimum reflection values in the 900 to 1700 nm range for juices (apple, grape, tomato) and jams (strawberry, peach, tomato) differed depending on the presence or absence of physical hazards. The presence of physical hazards was also confirmed by the color difference in the SAM image. The partial least squares discriminant analysis model (PLS-DA) and logistic regression implemented on the Jupyter Notebook platform through the Anaconda prompt provided accuracy, F1 score, specificity, and sensitivity based on the confusion matrix. The maximum value was 100.0 %, while the minimum value was 97.6 % of the result of the PLS-DA modeling in the testing set. Logistic regression modeling also had a similar result: the maximum value was 100.0 %, while the minimum value was 97.8 % in the testing set. The sensitivity value in the testing set, which is a meaningful result for detecting physical hazards, was a maximum of 100.0 % for both PLS-DA and logistic regression. Results from the current study suggest that the reflection value and SAM data obtained through hyperspectral imaging could build a big data platform for early determination of physical hazards during agricultural product processing.

Structural graph learning method for hyperspectral band selection

ABSTRACT Recently, graph learning-based hyperspectral band selection algorithms illustrate impressive performance for hyperspectral image (HSI) processing, whose goal is to select an optimal band combination containing less redundancy through the learned graph matrix. However, most of the previous methods work in single spectral domain, which neglects the rich image spatial information. Moreover, they typically model the graph matrix from a global perspective while ignoring the differences in spatial distribution that exist in diverse pixel and band regions. Based on the above considerations, to take full account of structure information in spatial and spectral domains, a structural graph learning method for hyperspectral band selection (SGLM) is designed. Specifically, SGLM constructs two matrices using the correlation between pixels and bands under the local perspective, which can capture the image structure information reasonably. Since the proposed method is modelled in both spatial and spectral dimensions, it is conducive to subsequent band subset generation task. Meanwhile, in order to guarantee local spatial consistency among bands, a Laplacian regularization term associated with the error matrix is introduced to the self-representation model. Additionally, considering that the importance of a band is consistent with its capability to reconstruct the entire band set, an adjacent bands reconstruction strategy is adopted to obtain the ultimate band combination, which can assess the significance of each band effectively. The comprehensive experimental analysis on four datasets demonstrates that the proposed SGLM model has outstanding superiority in comparison with several band selection algorithms.

Expression of Trichoderma spp. endochitinase gene improves red rot disease resistance in transgenic sugarcane.

Sugarcane (Saccharum spp.)is an economically useful crop grown globally for sugar, ethanol and biofuel production. The crop is vulnerable to fungus Colletotrichum falcatum known to cause red rot disease. The pathogen hydrolyses stalk parenchyma cells where sucrose is accumulated resulting in upto 75% losses in sugar recovery. In this study, transgenic sugarcane having resistance against red rot was developed by introducing Trichoderma spp. endochitinase following Agrobacterium mediated transformation. The transgene introduction and expression in genetically modified plants were verified through qRT-PCR revealing upto 6-fold enhancement in endochitinase expression than non-transgenic plants. Hyperspectral Imaging of transgenic plants displayed altered leaf reflectance spectra and vegetative indices that were positively correlated with ransgene expression. The bioassay with virulent pathotypes of C. falcatumCF08 and CF13 known for epiphytotic occurrence resulted in identification of resistant plant Chit 3-13.The plants with higher reflectance also displayed improved disease resistance, implying their early classification into resistant/susceptible. The losses in sucrose content were minimized (up to 4-fold) in inoculated resistant plant Chit 3-13 as compared to susceptible non-transgenic plant, and a fewer pathogen hyphae were detected in vascular cells of the former through optical microscopy. The electron micrographs confirmed sucrose-filled stalk parenchyma cells in Chit 3-13; in contrast, cells of non-transgenic inoculated plant were depleted of sucrose. The active sites involved in cleaving 1-4 β-glycoside bonds of N-acetyl-d-glucosaminein the pathogen hyphal walls were detected through endochitinase protein structural modelling. The transgenic sugarcane is an important source for in trogressingred rot resistance in plant breeding programs.

Feature extraction via 3-D homogeneous attribute decomposition for hyperspectral imagery classification

ABSTRACT Feature extraction is a core aspect in hyperspectral image classification, which can extract key information closely related to ground cover from complex scene, thus improving classification accuracy. Therefore, designing an effective feature extraction network is a hotspot and a challenge in the current research. In this paper, a feature extraction framework based on 3-D homogeneous attribute decomposition (3D-HAD) is proposed for HSI classification, which consists of the following key technologies. First, the principal component analysis algorithm is applied to the raw HSI to extract the principal components (PCs), and the raw HSI is clustered into many 3D superpixel blocks according to the first three PCs-based over-segmentation strategy. Then, a superpixel intrinsic attribute decomposition (SIAD) is designed to capture reflectance feature and suppress shading feature. Meanwhile, a metric entropy is introduced into the decomposition process to overcome the spectral-spatial weak assumption among pixels. Next, superpixel-guided recursive filtering is employed to preserve global details of HSI to enhance accuracy in HSI classification. Finally, the support vector machine classifier is used to obtain classification results of HSI. Experiments performed on several real hyperspectral data sets with limited training samples indicate that the proposed 3D-HAD method outperforms the classic, advance, and deep learning classification methods.

Compressed hyperspectral imaging based on image reflection intensity and differential fusion filtering

Existing spectral imaging technology based on compressed coding requires tens of minutes or even hours to obtain higher-quality spectral data. This limits their use in real dynamic scenarios and can only be discussed theoretically. Therefore, we propose a non-iterative algorithm model based on image reflection intensity-estimation aid (IRI-EA). The algorithm studies the approximate proportional relationship between the reflection strength of the RGB diagram and the corresponding spectrum image and reconstructs high-quality spectral data within about 20 s. By solving the difference map of the corresponding spectral scene, combining it with the spectral data of the IRI method, and introducing the total guidance (TG) filter, the reconstruction error can be significantly reduced, and the spectral reconstruction quality can be improved. Compared with other advanced methods, numerous experimental results indicate the advantages of this method in reconstruction quality and efficiency. Specifically, compared with the existing advanced methods, the average efficiency of our method has improved by at least 85%. Our reconstruction model opens up the possibility of processing real-time video and accelerating other methods.

Estimating the changes in mechanically expressible oil in terms of content and quality from ohmic heat treated mustard (Brassica juncea) seeds by Vis–NIR–SWIR hyperspectral imaging

Estimating the changes in mechanically expressible oil in terms of content and quality from ohmic heat treated mustard (Brassica juncea) seeds by Vis–NIR–SWIR hyperspectral imaging

Calibration verification of an underwater hyperspectral imaging push broom instrument to measure light in absolute units and field demonstration

The push broom design of an underwater hyperspectral imaging (UHI) instrument makes it possible to measure angle-resolved spectral radiance L(λ) in a plane. We describe the characterization of a commercial UHI instrument (UHI-4, Ecotone AS, Norway) and the spectral, geometric, and radiometric calibration transfer for measuring L(λ) in absolute units [µWcm−2nm−1sr−1]. We present a low-cost instrument characterization approach that is intended to be easily replicated for other users to perform their own calibration transfer. Cross-calibration with a RAMSES-ARC spectroradiometer (TriOS, Germany) in air and in water shows good linear correlation across the observed spectral range and over four orders of magnitude in dynamic range, with the UHI instrument providing higher sensitivity overall.

Fourier coded aperture transform rapid spectral imaging

The Fourier coded aperture transform hyperspectral imaging system (FCTS), which we proposed in our previous work, has excellent physical interpretability compared to other encoding methods. Due to the excellent energy concentration characteristic of Fourier encoding, it can achieve high-quality spectral imaging even under undersampling conditions. However, since the grayscale encoding method limits the system's sampling speed, we propose a binary sampling Fourier-encoded spectral imaging method. Besides, to address the significant errors of reconstructed spectral image at low sampling rates, we construct a spectral image enhancement network, which introduces spatial and channel attention modules to successively enhance the spatial and spectral information of the reconstructed spectral images. During the spectral image refinement and enhancement stage, band grouping is performed to alleviate the difficulty in feature extraction and make the training process more stable. The effectiveness of our proposed method has been validated on both public datasets and our own datasets. When the sampling times are reduced by an order of magnitude, on the three public datasets, the PSNR improved by 10.79 dB, the SSIM increased by 0.35, and the SAM decreased by 12.1° on average. On our own datasets, the PSNR improved by 12 dB, the SSIM increased by 0.24, and the SAM decreased by 6.5°.