Hyperspectral Imaging Images Research Articles

Hybrid hyperspectral image compression technique for non-iterative factorized tensor decomposition and principal component analysis: application for NASA’s AVIRIS data

Remote sensing data has shown tremendous potential for applications in various fields like land use mapping and detection, geologic mapping, water resource applications, wetland mapping, urban and regional planning, environment inventory, natural disaster assessment, archaeological applications, and others. Every day, thousands of gigabytes of memory are involved in capturing the hyperspectral remote sensing datasets. The compelling information present in these hyperspectral images (HSIs) is very minimal due to redundancy. Spatial and spectral correlations monopolize the acquired HSI data sets. Therefore, an algorithm that exploits these correlations and compresses the HSI tensors is proposed in this paper. First, the acquired HSI image (reflectance data) is subjected to the removal of geometric and radiometric errors. Second, spectral bands of interest affiliated to the underlying application are exclusively processed for principal component analysis (PCA). Results of this PCA are scrutinized to identify the absolute dependent components. Further, these components are exposed to a non-iterative factorized compression technique. As a result, HSI 3D tensors are disintegrated into 1D tensors. This tensor breakdown leads to a compression ratio as high as 3747:1 while the total encoding time observed is 332 s and RMSE is as low as 0.0017. Later, the original HSI is reconstructed back by the product of decomposed individual tensors and its PSNR is 53.03 dB. The proposed compression method targets the tucker decomposition-based HSI compression technique which is computationally complex and time consuming, and hence, a breakthrough is achieved with the technique introduced.

A Probabilistic Hyperspectral Imagery Restoration Method

Hyperspectral image (HSI) restoration is an important task of hyperspectral imagery processing, which aims to improve the performance of the subsequent HSI interpretation and applications. Considering HSI is always influenced by multiple factors—such as Gaussian noise, stripes, dead pixels, etc.—we propose an HSI-oriented probabilistic low-rank restoration method to address this problem. Specifically, we treat the expected clean HSI as a low-rank matrix. We assume the distribution of complex noise obeys a mixture of Gaussian distributions. Then, the HSI restoration problem is casted into solving the clean HSI from its counterpart with complex noise. In addition, considering the rank number need to be assigned manually for existing low-rank based HSI restoration method, we propose to automatically determine the rank number of the low-rank matrix by taking advantage of hyperspectral unmixing. Experimental results demonstrate HSI image can be well restored with the proposed method.

An efficient feature fusion in HSI image classification

In recent times, the fusion of spatial relaxation with spectral data has achieved remarkable success in target classification methods. Spatial relaxation is a scheme which exploits the neighbourhood relationship between the pixels of an image to minimize the spatio-spectral distortion. Application of spatial relaxation with spectral data can lead to reduce the noise effect and increase the class characterization. Such methods can also be applied to estimate the posteriors of the probabilistic classifier, to increase the classifier’s final accuracy. In this paper, we have introduced an edge based feature fusion method which helps in characterizing the class labels of hyperspectral image (HSI) in a better sense. It is an iterative method which exploits the spatial information from an image in such a manner that it assumes the feature preservation in vertical and horizontal directions for each pixel. With combining subspace regression based probabilistic method, the proposed method gives better accuracy for benchmark HSI datasets. Before this, we have implemented a fast Bayesian subspace regression method to achieve the posterior probabilities, for our edge feature relaxation method. Finally, we have compared the results with some recently proposed methods, and $$\alpha $$ expansion graph cut optimization method, which is an efficient technique to fuse the contextual knowledge in posterior probabilities.

An efficient approach for dimension selection and classification in HSI images

ABSTRACTIn this paper, we have proposed a divide and select approach for dimension selection (DSCS) and classification of hyperspectral images (HSI). The DSCS algorithm has been offered using a two-stage framework. In stage-1, we have generated the intermediate channels for every pair of spectral dimensions that are weakly correlated and highly significant. These channels have derived by splitting the adjacent dimensions without any loss of physical meaning. Intermediated channels are concatenated with original HSI in a better sense to transform the image into a more informative datacube. However, this leads to an increase in the dimensionality of the image. In stage-2, a trace and determinant based local feature response approach has applied to select the most informative dimensions of transformed HSI. We have exploited the local feature and scale selection methods to obtain significant channels. Finally, classification experiments have conducted for selected bands with SVM (Support Vector Machine) and LDA (Linear Discriminant Analysis). An expansion graph cut optimization has applied to improve the classification accuracy. This method has demonstrated as state-of-art for target detection in hyperspectral scenes.

Domain Transfer Learning for Hyperspectral Image Super-Resolution

A Hyperspectral Image (HSI) contains a great number of spectral bands for each pixel; however, the spatial resolution of HSI is low. Hyperspectral image super-resolution is effective to enhance the spatial resolution while preserving the high-spectral-resolution by software techniques. Recently, the existing methods have been presented to fuse HSI and Multispectral Images (MSI) by assuming that the MSI of the same scene is required with the observed HSI, which limits the super-resolution reconstruction quality. In this paper, a new framework based on domain transfer learning for HSI super-resolution is proposed to enhance the spatial resolution of HSI by learning the knowledge from the general purpose optical images (natural scene images) and exploiting the cross-correlation between the observed low-resolution HSI and high-resolution MSI. First, the relationship between low- and high-resolution images is learned by a single convolutional super-resolution network and then is transferred to HSI by the idea of transfer learning. Second, the obtained Pre-high-resolution HSI (pre-HSI), the observed low-resolution HSI, and high-resolution MSI are simultaneously considered to estimate the endmember matrix and the abundance code for learning the spectral characteristic. Experimental results on ground-based and remote sensing datasets demonstrate that the proposed method achieves comparable performance and outperforms the existing HSI super-resolution methods.

Fusion of hyperspectral and multispectral image by dual residual dense networks

The spatial resolution of hyperspectral image (HSI) is severely limited as a cost for higher spectral resolution. We propose a deep learning-based HSI super resolution method named dual residual dense networks (DRDNs) to overcome the resolution limitation by fusing low-resolution (LR) HSI and high-resolution (HR) multispectral image (MSI). The proposed model uses two symmetric subnets based on residual dense block to fully exploit deep features from HR-MSI and LR-HSI, respectively. Then the features are fused through a feature fusion subnet to generate the final reconstructed HR-HSI. Our DRDNs have realized end-to-end mapping from the original input of LR-HSI and HR-MSI to the desired HR-HSI directly. All the parameters are trained through a unified framework. Some experiments are also carried out to evaluate the performance of our proposed method. The results show that our proposed method gains significant improvement in simultaneously enhancing spatial resolution and preserving spectral consistency when compared to other HSI/MSI fusion methods proposed recently.

Adaptive Spatial-Spectral Feature Learning for Hyperspectral Image Classification

The combination of spectral and spatial information provides an effective way to improve the hyperspectral image (HSI) classification. However, the local spatial contexture changes with different neighborhood regions over the HSI image plane, and methods with fixed weights to integrate spatial information for all neighborhood regions could result in inaccurate spatial features, leading to adverse effects on classification performance. To address this issue, a novel adaptive spatial-spectral feature learning network (ASSFL) has been proposed to reflect spatial contexture changes and learn robust adaptive features in this paper. In the implementation of the proposed method, a convolution neural network (CNN) is first applied to learn weight features for each pixel within a hyperspectral patch and adaptive weights can be obtained based on a softmax normalization. Then, the shallow joint adaptive features can be acquired according to these weights. After that, a stacked auto-encoder (SAE) is proposed to further extract deeper hierarchical features for the final classification. The experimental results on four benchmark HSI data sets demonstrate that the proposed method can achieve competitive classification results compared with other existing classifiers.

Hyperspectral Tissue Image Segmentation Using Semi-Supervised NMF and Hierarchical Clustering.

Hyperspectral imaging (HSI) of tissue samples in the mid-infrared (mid-IR) range provides spectro-chemical and tissue structure information at sub-cellular spatial resolution. Disease states can be directly assessed by analyzing the mid-IR spectra of different cell types (e.g., epithelial cells) and sub-cellular components (e.g., nuclei), provided that we can accurately classify the pixels belonging to these components. The challenge is to extract information from hundreds of noisy mid-IR bands at each pixel, where each band is not very informative in itself, making annotations of unstained tissue HSI images particularly tricky. Because the tissue structure is not necessarily identical between the two sections, only a few regions in unstained HSI image can be annotated with high confidence, even when serial (or adjacent) hematoxylin and eosin stained section is used as a visual guide. In order to completely use both labeled and unlabeled pixels in training images, we have developed an HSI pixel classification method that uses semi-supervised learning for both spectral dimension reduction and hierarchical pixel clustering. Compared to the supervised classifiers, the proposed method was able to account for the vast differences in the spectra of sub-cellular components of the same cell type and to achieve an F1 score of 71.18% on twofold cross-validation across 20 tissue images. To generate further interest in this promising modality, we have released our source code and also showed that disease classification is straightforward after HSI image segmentation.

Self-Dictionary Regression for Hyperspectral Image Super-Resolution

Due to sensor limitations, hyperspectral images (HSIs) are acquired by hyperspectral sensors with high-spectral-resolution but low-spatial-resolution. It is difficult for sensors to acquire images with high-spatial-resolution and high-spectral-resolution simultaneously. Hyperspectral image super-resolution tries to enhance the spatial resolution of HSI by software techniques. In recent years, various methods have been proposed to fuse HSI and multispectral image (MSI) from an unmixing or a spectral dictionary perspective. However, these methods extract the spectral information from each image individually, and therefore ignore the cross-correlation between the observed HSI and MSI. It is difficult to achieve high-spatial-resolution while preserving the spatial-spectral consistency between low-resolution HSI and high-resolution HSI. In this paper, a self-dictionary regression based method is proposed to utilize cross-correlation between the observed HSI and MSI. Both the observed low-resolution HSI and MSI are simultaneously considered to estimate the endmember dictionary and the abundance code. To preserve the spectral consistency, the endmember dictionary is extracted by performing a common sparse basis selection on the concatenation of observed HSI and MSI. Then, a consistent constraint is exploited to ensure the spatial consistency between the abundance code of low-resolution HSI and the abundance code of high-resolution HSI. Extensive experiments on three datasets demonstrate that the proposed method outperforms the state-of-the-art methods.

Evaluation of growth characteristics of Aspergillus parasiticus inoculated in different culture media by shortwave infrared (SWIR) hyperspectral imaging

The growth characteristics of Aspergillus parasiticus incubated on two culture media were examined using shortwave infrared (SWIR, 1000–2500[Formula: see text]nm) hyperspectral imaging (HSI) in this work. HSI images of the A. parasiticus colonies growing on rose bengal medium (RBM) and maize agar medium (MAM) were recorded daily for 6 days. The growth phases of A. parasiticus were indicated through the pixel number and average spectra of colonies. On score plot of the first principal component [Formula: see text] and PC2, four growth zones with varying mycelium densities were identified. Eight characteristic wavelengths (1095, 1145, 1195, 1279, 1442, 1655, 1834 and 1929[Formula: see text]nm) were selected from PC1 loading, average spectra of each colony as well as each growth zone. Furthermore, support vector machine (SVM) classifier based on the eight wavelengths was built, and the classification accuracies for the four zones (from outer to inner zones) on the colonies on RBM were 99.77%, 99.35%, 99.75% and 99.60% and 99.77%, 99.39%, 99.31% and 98.22% for colonies on MAM. In addition, a new score plot of PC2 and PC3 was used to differentiate the colonies incubated on RBM and MAM for 6 days. Then characteristic wavelengths of 1067, 1195, 1279, 1369, 1459, 1694, 1834 and 1929[Formula: see text]nm were selected from the loading of PC2 and PC3. Based on them, a new SVM model was developed to differentiate colonies on RBM and MAM with accuracy of 100.00% and 99.99%, respectively. In conclusion, SWIR hyperspectral image is a powerful tool for evaluation of growth characteristics of A. parasiticus incubated in different culture media.

Hyperspectral Imaging and Classification for Grading Skin Erythema

Erythema is an inflammatory condition of the skin that is commonly used as a feature to monitor the progression of cutaneous diseases or treatment induced side effects. In radiation therapy, skin erythema is routinely assessed visually by an expert using standardized grading criteria. However, visual assessment (VA) is subjective and commonly used grading tools are too coarse to score the onset of erythema. Therefore, an objective method capable of quantitatively grading early erythema changes may help identify patients at higher risk for developing severe radiation induced skin toxicities. The purpose of this study is to investigate the feasibility of using hyperspectral imaging (HSI) for quantitative assessment of early erythema and to characterize its performance against VA documented on conventional digital photographic red-green-blue (RGB) images. Erythema was induced artificially on 3 volunteers in a controlled pilot study; and was subsequently measured using HSI, color imaging, and reflectance spectroscopy. HSI and color imaging data was analyzed using linear discriminant analysis (LDA) to perform classification. The classification results, including accuracy and precision, demonstrated that HSI is superior to color imaging in skin erythema assessment.

Application of hyperspectral imaging in detecting intracranial gliomas: An experimental study in mice

Objective To provide experimental foundation for application of hyperspectral imaging (HSI) in precise resection of the glioma and detection of its area and boundary in the mouse brain by HSI. Methods GL261 cells were implanted into the brain of C57 mice to generate the glioma. The images of 13 animal brains with gliomas (10 in vitro and 3 in vivo) were obtained by hyperspectral imaging and operating microscope. The spectrum of glioma and normal brain were extracted. The accuracy of HSI images were compared with that of 7.0-T MRI images. Results The HSI spectrum of tumor mass was significant different from normal brain tissue in the whole range of 460-700 nm. However, the HSI spectrum of infiltrating tumor margin (ITM) showed differences from internal capsule and corpus callosum in 460-600 nm, but no differences from dentate gyrus in the 460-700 nm range. Compared with the cortex and thalamus, the HSI spectral reflectance of ITM was lower in 500-600 nm range, but higher in 650-700 nm range. The mass and the boundary of gliomas could be clearly figured out in the HSI R700/R545 images. Based on the tumor areas calculated by hematoxylin-eosin staining (HE) staining, the accuracies of Red Green Blue (RGB) images, MRI T2, and HSI R700/R545 images were 81.9%±4.5%, 84.4%±4.7% and 92.4%±2.5%, respectively. Conclusion HSI can be used to identify the mass and boundary of mice gliomas accurately and serve as a potential real-time and label free imaging tool during precise glioma resection. Key words: Glioma; Magnetic resonance imaging; Neuronavigation; Hyperspectral imaging; Mice

Integrating spatial and spectral information for enhancing spatial features in the Gough map of Great Britain

Abstract The Gough map has been recognized as the earliest surviving map of Great Britain. The map, dated to late 14th or early 15th century, depicts many and sophisticated cartographical local details, which was unusual in European medieval maps. Moreover, the rewriting and re-inking of some names or areas could indicate the map was re-touched after its original production. The Gough map, today with restricted access and in a conservation environment, was displayed for some time at the Bodleian Library in Oxford, where it was frequently in contact with scholars. Despite this familiarization, the Gough map has not been widely researched and inquiries about the original purpose of the map or the context in which it was made remain unknown. Recent interest in the map has lead to the use of new technologies to image the map and produce sophisticated data that allows scholars and scientists to examine it taking advantage of the new information. In this paper, a technique for Pan-sharpening hyperspectral images (HSI) is introduced to the cultural heritage community. The Pan-sharpening process is applied specifically to HSI images of the Gough map of Great Britain with the purpose of using relatively low spatial resolution hyperspectral analysis techniques on its very fine features. The Pan-sharpening technique is based on Nearest-neighbor diffusion (NNDiffuse) and the spatial enhancing is aimed at faded features such as handwriting and some distinctive details that are only visible in high-resolution conservation photographs. It is shown here that the use of the NNDiffuse Pan-sharpening improves spatial features in HSI of historical artifacts without impacting the spectral fidelity, and that the findings in the analysis of these features could contribute to the understanding of the Gough map and its importance in the historical context of Great Britain.

Development of Noninvasive Classification Methods for Different Roasting Degrees of Coffee Beans Using Hyperspectral Imaging.

This study aimed to develop an approach for quickly and noninvasively differentiating the roasting degrees of coffee beans using hyperspectral imaging (HSI). The qualitative properties of seven roasting degrees of coffee beans (unroasted, light, moderately light, light medium, medium, moderately dark, and dark) were assayed, including moisture, crude fat, trigonelline, chlorogenic acid, and caffeine contents. These properties were influenced greatly by the respective roasting degree. Their hyperspectral images (874–1734 nm) were collected using a hyperspectral reflectance imaging system. The spectra of the regions of interest were manually extracted from the HSI images. Then, principal components analysis was employed to compress the spectral data and select the optimal wavelengths based on loading weight analysis. Meanwhile, the random frog (RF) methodology and the successive projections algorithm were also adopted to pick effective wavelengths from the spectral data. Finally, least squares support vector machine (LS-SVM) was utilized to establish discriminative models using spectral reflectance and corresponding labeled classes for each degree of roast sample. The results showed that the LS-SVM model, established by the RF selecting method, with eight wavelengths performed very well, achieving an overall classification accuracy of 90.30%. In conclusion, HSI was illustrated as a potential technique for noninvasively classifying the roasting degrees of coffee beans and might have an important application for the development of nondestructive, real-time, and portable sensors to monitor the roasting process of coffee beans.

Deformable Registration of Histological Cancer Margins to Gross Hyperspectral Images using Demons.

Hyperspectral imaging (HSI), a non-contact optical imaging technique, has been recently used along with machine learning technique to provide diagnostic information about ex-vivo surgical specimens for optical biopsy. The computer-aided diagnostic approach requires accurate ground truths for both training and validation. This study details a processing pipeline for registering the cancer-normal margin from a digitized histological image to the gross-level HSI of a tissue specimen. Our work incorporates an initial affine and control-point registration followed by a deformable Demons-based registration of the moving mask obtained from the histological image to the fixed mask made from the HS image. To assess registration quality, Dice similarity coefficient (DSC) measures the image overlap, visual inspection is used to evaluate the margin, and average target registration error (TRE) of needle-bored holes measures the registration error between the histologic and HSI images. Excised tissue samples from seventeen patients, 11 head and neck squamous cell carcinoma (HNSCCa) and 6 thyroid carcinoma, were registered according to the proposed method. Three registered specimens are illustrated in this paper, which demonstrate the efficacy of the registration workflow. Further work is required to apply the technique to more patient data and investigate the ability of this procedure to produce suitable gold standards for machine learning validation.

Heterospectral two-dimensional correlation analysis with near-infrared hyperspectral imaging for monitoring oxidative damage of pork myofibrils during frozen storage

Near-infrared (NIR) spectra contain abundant data, heterospectral two-dimensional correlation (H2D-CS) analysis offers a good way to interpret these data. For the first time, H2D-CS was used to correlate the NIR hyperspectral imaging (HSI) data with mid-infrared spectra and to identify feature-related wavebands for developing models for monitoring the oxidative damage of pork myofibrils during frozen storage. The HSI images were acquired at frozen state without thawing and the oxidative damage of myofibrils was assessed by carbonyl content. Results showed that the simplified PLSR model based on H2D-CS identified feature wavebands obtained determination coefficient in prediction (R2P) of 0.896 and root mean square error in prediction (RMSEP) of 0.177 nmol/mg protein, which was better than the partial least square regression (PLSR) model based on full wavebands (R2P = 0.856, RMSEP = 0.209 nmol/mg protein). Therefore, H2D-CS was effective in selecting feature-related wavebands of NIR HSI.

Label-free reflectance hyperspectral imaging for tumor margin assessment: a pilot study on surgical specimens of cancer patients.

.A label-free, hyperspectral imaging (HSI) approach has been proposed for tumor margin assessment. HSI data, i.e., hypercube , consist of a series of high-resolution images of the same field of view that are acquired at different wavelengths. Every pixel on an HSI image has an optical spectrum. In this pilot clinical study, a pipeline of a machine-learning-based quantification method for HSI data was implemented and evaluated in patient specimens. Spectral features from HSI data were used for the classification of cancer and normal tissue. Surgical tissue specimens were collected from 16 human patients who underwent head and neck (H&N) cancer surgery. HSI, autofluorescence images, and fluorescence images with 2-deoxy-2-[(7-nitro-2,1,3-benzoxadiazol-4-yl)amino]-D-glucose (2-NBDG) and proflavine were acquired from each specimen. Digitized histologic slides were examined by an H&N pathologist. The HSI and classification method were able to distinguish between cancer and normal tissue from the oral cavity with an average accuracy of , sensitivity of , and specificity of . For tissue specimens from the thyroid, the method achieved an average accuracy of , sensitivity of , and specificity of . HSI outperformed autofluorescence imaging or fluorescence imaging with vital dye (2-NBDG or proflavine). This study demonstrated the feasibility of label-free, HSI for tumor margin assessment in surgical tissue specimens of H&N cancer patients. Further development of the HSI technology is warranted for its application in image-guided surgery.

From Subpixel to Superpixel: A Novel Fusion Framework for Hyperspectral Image Classification

Supervised classification of hyperspectral images (HSI) is a very challenging task due to the existence of noisy and mixed spectral characteristics. Recently, the widely developed spectral unmixing techniques offer the possibility to extract spectral mixture information at a subpixel level, which can contribute to the categorization of seriously mixed spectral pixels. Besides, it has been demonstrated that the discrimination between different materials will be improved by integrating the geometry and structure information, which can be derived from the variance between neighboring pixels. Furthermore, by incorporating the spatial context, the superpixel-based spectral–spatial similarity information can be used to smooth classification results in homogeneous regions. Therefore, a novel fusion framework for HSI classification that combines subpixel, pixel, and superpixel-based complementary information is proposed in this paper. Here, both feature fusion and decision fusion schemes are introduced. For the feature fusion scheme, the first step is to extract subpixel-level, pixel-level, and superpixel-level features from HSI, respectively. Then, the multiple feature-induced kernels are fused to form one composite kernel, which is incorporated with a support vector machine (SVM) classifier for label assignment. For the decision fusion scheme, class probabilities based on three different features are estimated by the probabilistic SVM classifier first. Then, the class probabilities are adaptively fused to form a probabilistic decision rule for classification. Experimental results tested on different real HSI images can demonstrate the effectiveness of the proposed fusion schemes in improving discrimination capability, when compared with the classification results relied on each individual feature.

Characterization of myofibrils cold structural deformation degrees of frozen pork using hyperspectral imaging coupled with spectral angle mapping algorithm

The study investigated the possibility of using hyperspectral imaging (HSI) in the spectral range of 1000–2200nm to characterize myofibrils cold structural deformation degrees of frozen pork samples. The HSI images of pork samples frozen under different freezing rates were acquired in the frozen state without thawing. The myofibrils cold structural deformation degrees were evaluated by surface hydrophobicity (S0ANS) and Ca2+-ATPase activity. Spectral angle mapping (SAM) algorithm was used for the first time to extract the spectral information for regression. Compared with the optimized partial least square regression (PLSR) models based on selected wavebands by successive projections algorithm (SPA), the optimized PLSR models developed based on the spectral angles calculated by the SAM algorithm achieved comparable or even better performance with R2P of 0.896 for S0ANS and 0.879 for Ca2+-ATPase activity, respectively. The implications of the frozen meat spectrum were also analyzed in the current study.

An Improved Compression Algorithm for Hyperspectral Images based on DVAT-SVD

Hyperspectral image (HSI) compression has recently become a popular research area in remote sensing applications. However, existing clustering and compression techniques are not based on the intensity value and work by forming static rules. To overcome these issues, a novel lossy (nearly lossless) compression algorithm based on the distributed visual assessment cluster tendency (DVAT) – singular value decomposition (SVD) technique is proposed. At first, the given HSI image is preprocessed by cellular automata filtering to remove the white Gaussian and speckle noise. After that, the preprocessed image is split into bands, which are formed to a distance matrix. Here, the distances between the image bands are predicted to identify pixels with related intensity values. Hence, the DVAT technique is implemented to cluster the nearest distance images. Then, the label of the cluster is optimally selected for segregating the band into a band index. Consequently, the cluster index is formed and the image matrix is clustered based on the label features. After that, the SVD technique is applied to encode the HSI band image. Therefore, the discrete wavelet transform technique is applied for transformation and the transformed cell redundancy check technique is employed to encode the image band. Then, the image is reconstructed by reversing the above processes. The major advantages of the proposed method comprise a cleared clustered output, a correct index value, and the reduction of data losses during compression. During experimental results, the performance of the proposed method was evaluated in terms of the compression ratio, the mean-square error, and the peak signal-to-noise ratio.