SUSHI: An algorithm for source separation of hyperspectral images with non-stationary spectral variation

Hyperspectral image (HSI) classification has many applications in different diverse research fields. We propose a method for HSI classification using principal component analysis (PCA), 2D spatial convolution, and support vector machine (SVM). Our method takes advantage of correlation in both spatial and spectral domains in an HSI data cube at the same time. We use PCA to reduce the dimensionality of an HSI data cube. We then perform spatial convolution to the dimension-reduced data cube once and then to the convolved data cube for the second time. As a result, we have generated two convolved PCA output data cubes in a multiresolution way. We feed the two convolved data cubes to SVM to classify each pixel to one of the known classes. Experiments on three widely used hyperspectral data cubes (i.e., Indian Pines, Pavia University, and Salinas) demonstrate that our method can improve the classification accuracy significantly when compared to a few existing methods. Our method is relatively fast in terms of central processing unit computational time as well.

Hyperspectral (HS) images (HSI) provide a vast amount of spatial and spectral information based on the high dimensionality of the pixels in a wide range of wavelengths. A HS image usually requires massive storage capacity, which demands high compression rates to save space with preservation of data integrity. HS image can be deemed as three dimensional data cube where different wavelengths (W) form the third dimension along with X and Y dimensions. To get a better compression result, spatial redundancy of HS images can be exploited using different coders along X, Y, or W direction. This article focuses on taking maximum advantage of HS images redundancy by rearranging HS image into different 3D data cubes and proposes a directionlet based compression scheme constituted the optimal compression plane (OCP) for adaptive best approximation of geometric matrix. The OCP, calculated by the spectral correlation, is used to the prediction and determination of which reconstructed plane can reach higher compression rates while minimizing data loss of hyperspectral data. Moreover, we also rearrange the 3D data cube into different 2D image planes and investigate the compression ratio using different coders. The schema can be used for both lossless and lossy compression. Our experimental results show that the new framework optimizes the performance of the compression using a number of coding methods (inclusive of lossless/lossy HEVC, motion JPEG, JPG2K, and JPEG) for HSIs with different visual content.KeywordsHyperspectral imageslossless compressionoptimal compression plane (OCP)

Hyperspectral images (HSI) have high spectral and low spatial resolutions. However, multispectral images (MSI) usually have low spectral and high spatial resolutions. In various applications HSI with high spectral and spatial resolutions are required. In this paper, a new method for spatial resolution enhancement of HSI using high resolution MSI based on sparse coding and linear spectral unmixing (SCLSU) is introduced. In the proposed method (SCLSU), high spectral resolution features of HSI and high spatial resolution features of MSI are fused. In this case, the sparse representation of some high resolution MSI and linear spectral unmixing (LSU) model of HSI and MSI is simultaneously used in order to construct high resolution HSI (HRHSI). The fusion process of HSI and MSI is formulated as an ill-posed inverse problem. It is solved by the Split Augmented Lagrangian Shrinkage Algorithm (SALSA) and an orthogonal matching pursuit (OMP) algorithm. Finally, the proposed algorithm is applied to the Hyperion and ALI datasets. Compared with the other state-of-the-art algorithms such as Coupled Nonnegative Matrix Factorization (CNMF) and local spectral unmixing, the SCLSU has significantly increased the spatial resolution and in addition the spectral content of HSI is well maintained.

Spectroradiometric imaging is based on the fact that all materials display a specific spectral signature when light impinges on them. That is, they reflect, absorb, and emit electromagnetic waves in a particular pattern depending on the wavelength, leading to a spectrum for the material that can be used to classify it. An imaging spectroradiometer captures images where the spectral information of every pixel is collected for a set of contiguous discrete spectral bands. It is customary to classify these instruments as ultraspectral, hyperspectral, multispectral, and panchromatic imagers. Typically, ultraspectral data sets are made up of thousands of spectral bands, while hyperspectral sets have hundreds, multispectral tens, and panchromatic sets just a few. Thus, each ultraspectral or hyperspectral image contains large amounts of data which can be viewed as a cube with two spatial and one spectral dimension. Regardless of the application of the images, the analysis methods employed must deal with large quantities of data efficiently [1]. Originally, imaging spectroradiometers were used as airborne or spaceborne remote sensing instruments. During the last decade, spectral imaging has become a common advanced tool for remote sensing of the surface of Earth. It is being used to collect crucial data about many activities of economic or scientific relevance [2] as, for instance, mineral exploitation [3], agriculture and forestry use [4], environmental pollution [5], ecosystem studies [6], assessment of natural resources [7], water resource management [8], monitoring of coastal zones [9] and many others. For all these remote sensing applications the hyperspectral images obtained usually contain mixtures of spectra in every pixel due to the poor spatial resolution of the images when taken a large distance from the target. Thus a single pixel could typically correspond to one hundred or more square meters of land. As a result, most analysis methods developed had as their main objective to provide the segmentation of the images in terms of the ratio of endmembers present in every pixel to improve the spatial discrimination of these systems when analyzing different types of land covers [10].

Hyperspectral images with high spatial resolution play an important role in material classication, change detection, and others. However, owing to the limitation of imaging sensors, it is difcult to directly acquire images with both high spatial resolution and high spectral resolution. Therefore, the fusion of remotely-sensed images is an effective way to obtain high-resolution desired data, which is usually an ill-posed inverse problem and susceptible to noise corruption. To address these issues, a low-rank model based on tensor decomposition is proposed to fuse hyperspectral and multispectral images incorporating graph regularization, in which the logarithmic low-rank function is utilized to suppress the small components for denoising. Furthermore, this paper takes advantage of the local spatial similarity of remotely-sensed images to enhance the reconstruction performance by constructing spatial graphs, and also promotes signature smoothing between adjacent endmember spectra using the neighborhood-based spectral graph regularization. Finally, a set of efcient solvers is carefully designed via alternating optimization for closed-from solutions and computational reduction, in which vector-matrix operators are adapted to solve the three-dimensional core tensor. Experimental tests on several real datasets illustrate that the proposed fusion method yields better reconstruction performance than the current state-of-the-art methods, and can signicantly suppress noise at the same time.

Two long-wave infrared (LWIR) hyperspectral imagers have been under extensive development. The first one utilizes a microbolometer focal plane array (FPA) and the second one is based on an Mercury Cadmium Telluride (MCT) FPA. Both imagers employ a pushbroom imaging spectrograph with a transmission grating and on-axis optics. The main target has been to develop high performance instruments with good image quality and compact size for various industrial and remote sensing application requirements. A big challenge in realizing these goals without considerable cooling of the whole instrument is to control the instrument radiation. The challenge is much bigger in a hyperspectral instrument than in a broadband camera, because the optical signal from the target is spread spectrally, but the instrument radiation is not dispersed. Without any suppression, the instrument radiation can overwhelm the radiation from the target even by 1000 times. The means to handle the instrument radiation in the MCT imager include precise instrument temperature stabilization (but not cooling), efficient optical background suppression and the use of background-monitoring-on-chip (BMC) method. This approach has made possible the implementation of a high performance, extremely compact spectral imager in the 7.7 to 12.4 μm spectral range. The imager performance with 84 spectral bands and 384 spatial pixels has been experimentally verified and an excellent NESR of 14 mW/(m²srμm) at 10 μm wavelength with a 300 K target has been achieved. This results in SNR of more than 700. The LWIR imager based on a microbolometer detector array, first time introduced in 2009, has been upgraded. The sensitivity of the imager has improved drastically by a factor of 3 and SNR by about 15 %. It provides a rugged hyperspectral camera for chemical imaging applications in reflection mode in laboratory and industry.

In this paper, we propose a novel approach for spectral unmixing based on unsupervised learning using autoencoder with Inhomogeneous Gaussian Markov random field (IGMRF) as prior for regularization. The decoder part of our autoencoder has linear weights making it a linear mixture model (LMM). The weights represent the endmember matrix that makes the hidden unit of autoencoder as abundances. IGMRF is used to apply spatial regularization on abundances that also preserves the discontinuities. To incorporate the spectral regularization, we use IGMRF priors on endmembers. In addition, we also apply the spatial and spectral regularizations on the given hyperspectral images (HSI). IGMRF parameters at every pixel locations are calculated using initial estimates of endmembers and abundances. We obtain both endmembers and their abundances by optimizing the loss function that consists of a data term and IGMRF prior terms. Experiments are performed on different noise level synthetic data and two real data (Jasper ridge and Urban). The results of the proposed approach are better when compared to existing state-of-the-art approaches.

In this paper, we propose a frequency-domain approach for convolutive blind separation of nonstationary acoustic sources (ConvBSS). We focus on the challenging problem of permutation ambiguity correction and introduce a new method using spectrum characterization of acoustic signals. The proposed approach involves the computation of the harmonic product spectrum (HPS) to take advantage of the sourcespsila time-frequency representation. The permutation ambiguity problem is then resolved by searching the permutation that maximizes the HPS-based correlation between the constructed harmonic spectrum and the estimated one at each resolution step. We demonstrate through experiment simulations that the ConvBSS is achieved in real environments.

Hyperspectral image (HSI) has shown promising results in many fields because of its high spectral resolution. However, redundancy and noise in spectral dimension seriously affect the classification of HSI. For this reason, many popular dimensionality reduction (DR) methods are proposed to solve the problem. The local discriminant embedding (LDE) as an effective non-linear method for DR can be more discriminative by constructing two neighborhood graphs. However, HSI is very easy influenced by noise, and the LDE algorithm based on K nearest neighborhood is highly susceptible to interference from extreme point, which may lead to inaccurate graph construction and poor performance of classification. To overcome the problem and retain the advantages of LDE, a modified local discriminant embedding (MLDE) is firstly applied on HSI by constructing neighborhood graphs on a new spectral feature space instead of the original space. We use variance to characterize the pixels similarity of the same class and use covariance to characterize the separation of different classes of pixels. The combination of variance and covariance makes pixels in the same class to be closer and makes greater separation of pixels from different classes, which enhances classification performance of HSI. The way of representing data by using variance and covariance can attenuate the effects of noise. The Log-Euclidean metric is used to capture the similarity between spectral vectors, which can provide a more accurate similarity evaluation than euclidean distance. The experimental results of two hyperspectral datasets demonstrate the effectiveness of our proposed MLDE method.

Spectrum sharing between radar and communication systems has attracted substantial recent research attention. In order to avoid the mutual interference, source separation of the aliasing signal is a promising solution for joint communication and radar (JCR) sensing. However, most of the existing source separation methods are very sensitive to wide-band signals with rapidly varying instantaneous amplitude (IA) and not applicable to scenarios with complex overlapping subsignals. To overcome the above limitations, a data-driven sparsity-based source separation method is proposed. Without depending on signal parameters, a sparse signal observation model of the received signal is established. Then, source separation is transformed into estimation of the IA and instantaneous frequency (IF), and an iterative method is developed. Moreover, the Cramér-Rao lower bound (CRLB) of the source separation error (SSE) variance is derived based on the CRLB of IA and IF. Simulation results demonstrate that our proposed method is superior to the benchmark schemes and is robust to noise and initial IF errors.

Advances in hyper spectral imaging sensors makes it possible to obtain high dimensional and high resolution data which poses challenges for computational cost and time. As the inter-band resolution is high in hyper spectral images, the spectral dimension could be reduced without significant loss of useful information. This requires careful application of dimensionality reduction techniques to reduce the dimension but preserve the original spectral characteristics. It is observed in the literature that filtering techniques can address some of the challenges concerning dimensionality reduction in hyper spectral imaging. In this paper a dimensionality reduction method adopting Discrete Wavelet Transform and Distance Classifier Correlation filter has been proposed for hyper spectral imaging. The experimental results show that the combination of ID-DWT with DCCF outperforms the other methods in dimensionality reduction and classification of given data set.

Face recognition is still a challenging problem as there is a high possibility that the differences existing within a person or subject can exceed the differences present between different persons. Most of the current research work in the field of biometric face recognition has dealt with (Red-GreenBlue) RGB images. However, Hyper Spectral Imaging (HSI), introduces the spectral dimension for improved discrimination and leads to building a more efficient face recognition system. The spectral dimension adds more intricate details to the image. As a result hyper spectral imaging helps in improving face recognition accuracy. The faces are captured at varying spectral wavelengths of the electromagnetic spectra. Hyper spectral imaging often increases facial discrimination by capturing more biometric measurements and thus revealing information that is not revealed by the commonly used RGB images. The proposed methodology for face recognition using hyper spectral imaging includes performing band selection and band fusion on the hyper spectral face cubes and then classifying them using 3 Dimensional Convolution Neural Networks (3D-CNN). The classification using 3D-CNN gave a promising accuracy of 97.3%.

A multimodal method for skin cancer diagnosis based on Dermatoscopy (DS) image analysis, Hyperspectral Imaging (HSI), Raman/Autofluorescense analysis (RS/AF) and Optical Coherence Tomography (OCT) is presented. A universal multi-class classifier for OCT is described. The diagnostic accuracy near 90% (depended from tissue type) has been demonstrated for in vivo and ex vivo cases. A multimodal unit for diagnosis of cutaneous malignant neoplasms has been proposed. It include following pure optical modalities: Raman Spectroscopy (NIR, 785 nm) in combination with Autofluorescence analysis (485 and 785 nm), Optical Coherence Tomography (840±45 nm) and Hyperspectral Imaging (visible range, 450-750 nm, 2.3 nm spectral resolution). The method is also supplemented with a Multispectral Digital Dermatoscopy, which is a perfect solution for preliminary screening and diagnosis of pathological areas by general physician. The own developed DS tool is based on the Basler acA1920-25uc camera (RGB, 12 bit/px, 1920*1080), its resolution on the surface of the skin is about 13 μm/px. The device body has been designed and printed on a 3D printer. Tissue lighting is carried out by various LEDs: UV 365 nm, White 4000 K (polarized and unpolarized LEDs); Blue 465-485 nm; Green 520-535 nm; Red 620-630 nm LEDs. The results of several in vivo and ex vivo series of experiments are presented. In vivo studies were performed in Samara Oncology Clinical Center. A total of 220 patients were examined with Malignant Melanoma (MM), Basal Cell Carcinoma (BCC) and other benign and malignant tumors. 232 tissue samples surgically resected from patients were examined ex vivo. All samples diagnoses were confirmed by histology reports after resection. A final accuracy of 97.3% was obtained using RS/AF analysis. Moreover, for case of Malignant/Benign in vivo case, it reached up to 100% with low-cost Ocean Optics QE65 Pro spectrograph and PLS analysis. A final separation of classes for OCT ex vivo leads to a high (95%) accuracy for two-classes cases and 75% for the four classes case. Combination of 2D and 3D OCT data give us an opportunity to receive common information about tumor (geometric and morphological characteristics) from data cubes and use more powerful algorithms for counting features (fractal, textural, geometric) on separated scans. These categories of features provide closer connection with ABCDE criteria (asymmetry, borders irregularity, color, diameter, evolution). Final accuracies 90% for in vivo HSI is very similar to in vivo DS accuracy, which reaches 93%. The optical density spectral feature has been used as a main diagnostic characteristic for HSI and topology texture and color analysis for RGB DS.

Hyperspectral imaging (HSI) is a powerful tool widely used for various scientific and industrial applications due to its ability to provide rich spatiospectral information. However, in exchange for multiplex spectral information, its image acquisition rate is lower than that of conventional imaging, with up to a few colors. In particular, HSI in the infrared region and using nonlinear optical processes is impractically slow because the three-dimensional (3D) data cube must be scanned in a point-by-point manner. In this study, we demonstrate a framework to improve the spectral image acquisition rate of HSI by integrating time-domain HSI and compressed sensing. Specifically, we simulated broadband coherent Raman imaging at a record high frame rate of 25 frames per second (fps) with 100 pixels × 100 pixels, which is 10 × faster than that of previous work, based on an experimentally feasible sampling scheme utilizing 3D Lissajous scanning.

In previous studies, the keypoints and their feature descriptors of a hyperspectral image have not received enough attention. Compared with RGB images, the rich feature information of a hyperspectral image can not only constrain the spatial positions of the keypoints, but also provide rich spectral features for their descriptors. In this paper, we propose a deep neural network method called Hyperspectral SuperPoint Network (HSSPN) to detect the keypoints from a hyperspectral image and generate the corresponding feature descriptors. The HSSPN method contains a Shared Encoder (which is responsible for fully extracting the features of the spatial and spectral dimensions) and two decoders: the Keypoint Decoder accurately detects the keypoints according to the feature map outputted by the Shared Encoder, while the Descriptor Decoder is responsible for generating stable feature descriptors. We collected some hyperspectral images with an electronically controlled turntable, and then annotated them with a Hyperspectral Homographic Adaptation (HHA) network. The trained HSSPN gave an excellent keypoint matching performance on the test set. Hyperspectral 3D reconstruction can obtain information in a 3D space and a spectral space at the same time. Most of the previous reconstruction methods are separated for the treatments of 3D reconstruction and hyperspectral imaging, which require complicated registrations. In this study, we propose a SfM-based hyperspectral 3D reconstruction method, in which we first collect hyperspectral images from multiple viewing angles, and then apply the HSSPN method to detect the keypoints and calculate their feature descriptors. These keypoints and feature descriptors are then used to reconstruct each spectral image of the target with the SfM method. The 3D reconstruction models at these wavelengths are simply superimposed to obtain a hyperspectral 3D reconstruction model. Compared with conventional SIFT-based SfM methods, our method not only contains richer spectral information, but also achieves better results in the reconstruction accuracy.