Sparse representation theory for classification is an active research area. Signals can potentially have a compact representation as a linear combination of atoms in an overcomplete dictionary. In this paper, a novel classification method is proposed, which combines sparse-representation-based classification (SRC) and K-nearest neighbor classifier for remote sensing image. Based on the extracted multidimensional features which are used to constitute an overcomplete dictionary, the image is expressed as the product of the dictionary and coefficient of sparse representation. Then the test image is reconstructed by utilizing correlation and distance information between the image and each class simultaneously. Finally, each image will be assigned a class label based on minimizing the reconstruction error. And then, the proposed method has been extended to a kernelized variant to solve linearly inseparable problems. The experimental results show that the proposed method and its variant not only improve the classification performance over SRC but also outperform typical classifiers, such as support vector machine(SVM), especially when the number of training samples is limited.
Hyperspectral anomaly detection, as an important application of hyperspectral remote sensing, is widely used in mineral exploration, environmental monitoring, military reconnaissance, etc. The anomalies in hyperspectral image mainly refer to the spectral anomalies caused by the reflection or radiation of some special objects. They have the characteristics of small scales, no prior information, and low probability of occurrence. Such anomalies in the military, agriculture, geology and other fields often contain much important information and has great application value. The sparsity score estimation algorithm is a global algorithm with high stability but as well as high false alarm rate. In this paper, we propose a hyperspectral anomaly detection algorithm based on non-negative sparsity score estimation. Firstly, the initial dictionary is obtained by K-SVD algorithm. After the sparse representation model of hyperspectral image is generated by orthogonal matching pursuit algorithm, dictionary atoms and corresponding coefficients are updated through singular value decomposition of the error term, then the optimization is achieved by successive iteration. Secondly, the nonnegative constraint condition is introduced into the sparse representation model, and the non-negative sparse coefficient is solved by the non-negative sparse coding. Finally, the non-negative sparse coefficients are used to calculate the atom usage probability, so as to infer whether the corresponding image pixels are anomaly or not. The experiments conducted on hyperspectral images show that the proposed method is superior to some typical methods, which has a lower false alarm rate while remains the 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.
The hyperspectral image in thermal infrared domains provide information, such as temperature and emissivity, about different kinds of materials. These information can be used for a wide number of applications such as mineral mapping, bathymetry, indoor and outdoor detection of chemicals. But because of the limitation of spatial resolution and the characteristics of thermal infrared sensor, there are many mixed pixels in the data, whose temperature,emissivity and abundance of different components can be hard to estimate. In this paper, a new method to estimate the parameters in pure and mixed pixels is proposed based on linear and nonlinear optimization. Firstly, the standard temperature and emissivity separation (TES) algorithm is applied on pure pixels of different materials selected by supervise or unsupervised methods to get the initial temperature. Secondly, the emissivity in different bands can be retrieved by minimizing the reconstruction error, which the more accurate temperature is optimized with. The emissivity in one band is trained by the samples in the same band but in different pixels, while the temperature is trained by different bands in one pixel. Lastly, the abundance and temperature of components in mixed pixels are estimated based on a linear mixture model of the bottom of atmosphere radiance as full constraint linear optimization problem and nonlinear optimization problem. The method is also analyzed with respect to sensitivity to the noise and different parameters’ influences on estimation errors.
Classification of real-world remote sensing images is a challenging task because of complex spectral–spatial information with high-dimensional feature vectors. Most of the traditional classification approaches directly treat data as vectors, which usually results in a small sample size problem and abundant redundant information; thus, they inevitably degrade the performance of the classifier. To overcome the drawbacks, we take advantage of the benefits of local scatters and tensor representation and propose a framework for hyperspectral image (HSI) classification through combining local tensor discriminant analysis (LTDA) with spectral–spatial feature extraction. First, we use a well-known spectral–spatial feature extraction approach to extract abundant spectral–spatial features as feature tensors. Then, based on class label information, LTDA is used to eliminate redundant information and to extract discriminant feature tensors for the subsequent classification. Two real HSIs are used as experimental datasets. The obtained results indicate that the proposed method exhibits good performance, while using a small number of training samples.
KEYWORDS: Hyperspectral imaging, RGB color model, Signal processing, Detection and tracking algorithms, Statistical modeling, Data modeling, Vegetation, Mathematical modeling, Information technology, Minerals
Current algorithms of endmember extraction generally need to determine the number of endmembers manually. However, the number of endmembers is unknown in practical application, so an automated and iterative endmember extraction algorithm is put forward in this paper to solve the problem. Firstly, due to the spectral information of endmember is similar with its neighbors but noise is independent with others, we analyze the relevance between pixels and endmember in the concentric sliding window centered at each test endmember in order to eliminate the influence of noise. Then, due to the independence among endmembers, a candidate set formed of endmembers which have been extracted is constructed. We compute the correlation between the new endmember and the candidates in the set each time, if the largest correlation is small; the new one is added to the set. If the new one fails to join the set directly, we can take it to replace the existed in the set to increase the distance among endmembers. Finally, if the endmembers in the set remain unchanged in a few times, the iteration stops. The experiment shows that the improved algorithm have a near accuracy of endmember extraction with the traditional algorithm, meanwhile it weakens the influence of noise on the endmember extraction.
In hyperspectral image processing, anomaly detection is a valuable way of searching targets whose spectral characteristics are not known, and the estimation of background signals is the key procedure. On account of the high dimensionality and complexity of hyperspectral image, dimensionality reduction and background suppression is necessary. In addition, the complementarity of different anomaly detection algorithms can be utilized to improve the effectiveness of anomaly detection. In this paper, we propose a novel method of anomaly detection, which is based on clustering of optimized K-means and decision-level fusion. In our proposed method, pixels with similar features are firstly clustered using an optimized k-means method. Secondly, dimensionality reduction is conducted using principle component analysis to reduce the amount of calculation. Then, to increase the accuracy of detection and decrease the false-alarm ratio, both Reed-Xiaoli (RX) and Kernel RX algorithm are used on processed image. Lastly, a decision-level fusion is processed on the detection results. A simulated hyperspectral image and a real hyperspectral one are both used to evaluate the performance of our proposed method. Visual analysis and quantative analysis of receiver operating characteristic (ROC) curves show that our algorithm can achieve better performance when compared with other classic approaches and state-of-the-art approaches.
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.