Band selection (BS) algorithms are an effective means of reducing the high volume of redundant data produced by the hundreds of contiguous spectral bands of Hyperspectral images (HSI). However, BS is a feature selection optimization problem and can be a computationally intensive to solve. Compressive sensing (CS) is a new minimally lossy data reduction (DR) technique used to acquire sparse signals using global, incoherent, and random projections. This new sampling paradigm can be implemented directly in the sensor acquiring undersampled, sparse images without further compression hardware. In addition, CS can be simulated as a DR technique after an HSI has been collected. This paper proposes a new combination of CS and BS using band clustering in the compressively sensed sample domain (CSSD). The new technique exploits the incoherent CS acquisition to develop BS via a CS transform utilizing inter-band similarity matrices and hierarchical clustering. It is shown that the CS principles of the restricted isometric property (RIP) and restricted conformal property (RCP) can be exploited in the novel algorithm coined compressive sensing band clustering (CSBC) which converges to the results computed using the original data space (ODS) given a sufficient compressive sensing sampling ratio (CSSR). The experimental results show the effectiveness of CSBC over traditional BS algorithms by saving significant computational space and time while maintaining accuracy.
Recently, many spectral-spatial hyperspectral image classification techniques have been developed, such as widely used EPF-based and composite kernel-based approaches. However, the performance of these types of spectral-spatial approaches are generally depends on both techniques and its guided spatial feature information. To address this issue, an unsupervised subpixel detection based hyperspectral feature extraction for classification approach is proposed in this paper. Harsany-Farrand-Chang (HFC) method is utilized to estimate the number of distinct features of hyperspectral image can be decomposed into, and simplex growing algorithm (SGA) is utilized to generate endmembers as initial condition for K-means clustering. Subpixel detection maps are generated by constrained energy minimization (CEM) using centroid of K-means clusters. To capture spatial information, multiple Gaussian feature maps are generated by applying Gaussian spatial filters with different on CEM detection maps, and PCA is used to reduce the dimension of multiple Gaussian feature maps, and feedback it into hyperspectral band images to reprocess K-means in an iteration manner. The proposed unsupervised approach is evaluated by supervised approaches such as iterative CEM (ICEM), EPF-based, and composite kernel-based methods, and results shows that most classification performance is improved.
In hyperspectral image classification, how to jointly take care of spectral and spatial information received considerable interest lately, and many spectral-spatial classification approaches have been proposed. Unlike spectral-spatial classifications which are developed from traditional aspect, iterative constrained energy minimization (ICEM) and iterative target-constrained interference-minimized classifier (ITCIMC) approaches are developed from subpixel detection and mixed pixel classification point of view, and generally performs better than existing spectral-spatial approaches in terms of several measurements, such as accuracy rate and precision rate. Recently, convolutional neural networks (CNNs) have been successfully applied to visual imagery classification and have received great attention in hyperspectral image classification, due to the outstanding ability of CNN to capture spatial information. This paper extends ICEM to iterative constrained energy minimization convolution neural network approach for hyperspectral image classification. In order to capture spatial information, instead of Gaussian filter, CNN is utilized to generate binary pixelwise classification map for constrained energy minimization (CEM) detection results, and CNN classification map is feedbacked into hyperspectral bands, and then CEM detection is reprocessed in an iteration manner. Since CNN can reduce the performance of precision rate, a background recovery procedure is designed, to recover background detection map from CEM detection map and add it into CEM result as a new detection map.
Owing to significant geometric distortions and illumination differences, high precision and robust matching of multisource remote sensing image registration poses a challenge. This paper presents a new approach, called iterative scale invariant feature transform (ISIFT) with rectification (ISIFTR), to remote sensing image registration. Unlike traditional SIFT-based methods or modified SIFT-based methods, the ISIFTR includes rectification loops to obtain rectified parameters in an iterative manner. The SIFT-based registration results is updated by rectification loops iteratively and terminated by an automatic stopping rule. ISIFTR works in three stages. The first stage is used to capture consistency feature sets with maximum similarity followed by a second stage to compare the registration parameters between two successive iterations for updating and finally concluded by a third stage to terminate the algorithm. The experimental results demonstrate that ISIFTR performed better registration accuracy than SIFT without rectification. By comparing the iteration curve based on the four different similarity metric, the results illustrate that the RIRMI-based rectification obtains better results than other similarity metrics.
Unsupervised target generation for hyperspectral imagery (HSI) have generated great interest in the hyperspectral community. However, most of the current unsupervised target generation algorithms have to process large HSI data, which is acquired using the traditional Nyquist-Shannon sampling theorem, resulting in data with high band-to-band correlation. As a consequence, these algorithms end up processing redundant information, raising the demand for large memory storage, processing time, and transmission bandwidth. In the past, some efforts have been dedicated to dealing with the redundant information via data reduction (DR) or data compression post-acquisition. However, to the best of our knowledge, this challenge has been addressed outside the context of Compressive Sensing (CS). This paper applies CS data acquisition process at the sensor level so that the redundant information is removed at the early stage of the data processing chain. The main advantage of our approach is that it employs a random sensing process, and the concept of universality, to randomly sense the HSI bands and produce data containing the bare minimum information. We take advantage of CS Restricted Isometric Properties (RIP), Restricted Conformal Properties (RCP), and newly derived orthogonal sub-space projection (OSP) properties to perform automatic target generation process (ATGP) in the compressively sensed band domain (CSBD), instead of in the original data space (ODS), where the HSI data contains full spectral bands. Our experimental results show that, by working in the CSBD, we avoid processing redundant data and still maintain performance results that are comparable with the performance results obtained in the ODS.
Tellurium dioxide is the most widely used uniaxial crystal for acousto-optic devices. Acousto-optic tunable filters based on this material can cover spectral range from UV to MWIR in a non-collinear configuration. The diffracted narrow band output beams have orthogonal linear polarizations, propagating in different directions, allowing the filter to act as polarizing beam splitter/analyzer as well. To achieve full electronic tuning, two liquid crystal variable retarders are used to measure all six polarization states used in the calculation of Stokes vector. We will present the design of the instrument, test results, and performance considerations.
Magnetic Resonance (MR) images can be considered as multispectral images so that MR imaging can be processed by
multispectral imaging techniques such as maximum likelihood classification. Unfortunately, most multispectral imaging
techniques are not particularly designed for target detection. On the other hand, hyperspectral imaging is primarily
developed to address subpixel detection, mixed pixel classification for which multispectral imaging is generally not
effective. This paper takes advantages of hyperspectral imaging techniques to develop target detection algorithms to find
lesions in MR brain images. Since MR images are collected by only three image sequences, T1, T2 and PD, if a
hyperspectral imaging technique is used to process MR images it suffers from the issue of insufficient dimensionality.
To address this issue, two approaches to nonlinear dimensionality expansion are proposed, nonlinear correlation
expansion and nonlinear band ratio expansion. Once dimensionality is expanded hyperspectral imaging algorithms are
readily applied. The hyperspectral detection algorithm to be investigated for lesion detection in MR brain is the well-known
subpixel target detection algorithm, called Constrained Energy Minimization (CEM). In order to demonstrate the
effectiveness of proposed CEM in lesion detection, synthetic images provided by BrainWeb are used for experiments.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.