Segmented Mixture-of-Gaussian Classification for Hyperspectral Image Analysis

Abstract

Abstract-The same high dimensionality of hyperspectral imagery that facilitates detection of subtle differences in spectral response due to differing chemical composition also hinders the deployment of traditional statistical pattern-classification procedures, particularly when relatively few training samples are available. Traditional approaches to addressing this issue, which typically employ dimensionality reduction based on either projection or feature selection, are at best suboptimal for hyperspectral classification tasks. A divide-and-conquer algorithm is proposed to exploit the high correlation between successive spectral bands and the resulting block-diagonal correlation structure to partition the hyperspectral space into approximately independent subspaces. Subsequently, dimensionality reduction based on a graph-theoretic localitypreserving discriminant analysis is combined with classification driven by Gaussian mixture models independently in each subspace. The locality-preserving discriminant analysis preserves the potentially multimodal statistical structure of the data, which the Gaussian mixture model classifier learns in the reduced-dimensional subspace. Experimental results demonstrate that the proposed system significantly outperforms traditional classification approaches, even when few training samples are employed.

Spectral–Spatial Classification of Hyperspectral Images With a Superpixel-Based Discriminative Sparse Model

“…Generally, although the aforementioned classifiers can effectively utilize the spectral information of the HSI, they do not consider the spatial context. Recently, to further improve classification performance, approaches based on composite kernels [9], support vector conditional random fields [10], and segmentation [11], [12], have been proposed to incorporate spatial information into the analysis of HSIs. In addition, some other recent classification approaches have focused on the design of effective feature extraction techniques (e.g., clonal selection feature-selection [13], extended morphological profiles [14], tensor discriminative locality alignment [15] and multiple features combination [16], [17]).…”

Compressive data fusion for multi-sensor image analysis

“…This approach is clearly suboptimal, thereby underscoring the motivation for performing at-sensor compressive measurements prior to reconstruction or image analysis. Fuse-PD refers to an approach wherein we employ our classification strategy individually in the compressivemeasurement domain of each source, and fuse posterior probabilities through a logarithmic opinion pool (LOGP) [18,19]. Finally, Compressive Fusion connotes compressive data fusion of incoming compressive measurements, followed by a single Bayesian classifier.…”

Robust joint sparsity model for hyperspectral image classification

“…which is similar to the JSRC model discussed in section 2.2 and also could be solved by the SOMP algorithm [25]. For problem (13), the optimization with respect to S (k+1) is formulated by…”

“…A large number of HSI classification methods have been proposed, based on artificial neural networks [4], multinomial logistic regression [5], [6] and support vector machines (SVM) [7], just to name a few. With the target of exploiting spatial information in the classification task, spatial-spectral classification approaches have been developed, including SVM with composite kernels [8], methods based on mathmatical morphology [9][10][11][12] and image segmentation [13].…”