Exploiting manifold geometry in hyperspectral imagery

Bachmann, Charles M.; Ainsworth, T.L.; Fusina, R.A.

doi:10.1109/tgrs.2004.842292

Cited by 398 publications

(211 citation statements)

References 33 publications

Supporting

Mentioning

208

Contrasting

Unclassified

Order By: Relevance

“…High-dimensional hyperspectral data usually lie on certain low-dimensional manifold structures [22], [23]. Remote sensing data of a given class typically occur in spatially contiguous clusters; thus, the image spatial space can be viewed as the most natural "low-dimensional manifold space."…”

mentioning

confidence: 99%

Active Learning via Multi-View and Local Proximity Co-Regularization for Hyperspectral Image Classification

Wei

Crawford

2011

IEEE J. Sel. Top. Signal Process.

View full text Add to dashboard Cite

Abstract-A novel co-regularization framework for active learning is proposed for hyperspectral image classification. The first regularizer explores the intrinsic multi-view information embedded in the hyperspectral data. By adaptively and quantitatively measuring the disagreement level, it focuses only on samples with high uncertainty and builds a contention pool which is a small subset of the overall unlabeled data pool, thereby mitigating the computational cost. The second regularizer is based on the "consistency assumption" and designed on a spatial or the spectral based manifold space. It serves to further focus on the most informative samples within the contention pool by penalizing rapid changes in the classification function evaluated on proximally close samples in a local region. Such changes may be due to the lack of capability of the current learner to describe the unlabeled data. Incorporating manifold learning into the active learning process enforces the clustering assumption and avoids the degradation of the distance measure associated with the original high-dimensional spectral features. One spatial and two local spectral embedding methods are considered in this study, in conjunction with the support vector machine (SVM) classifier implemented with a radial basis function (RBF) kernel. Experiments show excellent performance on AVIRIS and Hyperion hyperspectral data as compared to random sampling and the state-of-the-art SVM SIMPLE .

show abstract

mentioning

confidence: 99%

Active Learning via Multi-View and Local Proximity Co-Regularization for Hyperspectral Image Classification

Wei

Crawford

2011

IEEE J. Sel. Top. Signal Process.

View full text Add to dashboard Cite

show abstract

“…There are multiple sources of nonlinearity, one of the more significant sources, especially in land-cover classification application, stems from wavelength dependent nonlinear reflectance defined by the bidirectional reflectance distribution function (BRDF). Another source of nonlinearity arises from complex scattering of energy in vegetation and nonlinear attenuation of energy in water bodies [5].…”

Section: Introductionmentioning

confidence: 99%

“…These methods are developed to represent high dimensional nonlinear phenomena in lower dimensional spaces, the imbedded features have been used to address the nonlinearity in hyperspectral data; they are potentially useful for classification of hyperspectral data. Bachmann et al [5]demonstrated Isomap potential for data representation and classification of hyperspectral data. The problem of heavy computational load for large-scale remote sensing data sets was tackled by a scalable approach based on various methods for aligning manifolds derived from image subsets, experiments showed that features extracted by Isomap were able to explain a greater amount of variance in the data than MNF coordinates; The proposed ENH-Isomap was used for bathmetric retreival, and produced retrieval errors for spatially and temporally disjoint test sets that were comparable to those of bathymetric LiDAR [10].…”

Section: Introductionmentioning

confidence: 99%

Subspace Feature Analysis of Local Manifold Learning for Hyperspectral Remote Sensing Images Classification

Ding¹,

Tang²,

Li³

2014

Appl. Math. Inf. Sci.

View full text Add to dashboard Cite

Dimensionality reduction and segmentation have been used as methods to reduce the complexity of the representation of hyperspectral remote sensing images. In this study, a new object-oriented mapping approach is proposed based on nonlinear subspace feature analysis of hyperspectral remote sensing images. Nonlinear local manifold learning approaches for feature extraction were utilized to obtain subspace feature representation of hyperspectral remote sensing images. Afterwards, with a proper selection of parameters, the extracted subspace feature images were fed into the object-oriented system. Texture features derived from gray level co-occurrence matrix and wavelet filter with the use of SVM classifier at the pixel level of the feature images were also used to evaluate the proposed algorithm. Experiments are conducted on the AVIRIS dataset with 220 spectral bands, covering an agricultural area. Classification results show that the proposed object-oriented subspace analysis approach can give significantly higher accuracies than the traditional pixel-level and texture-based subspace feature classification.

show abstract

“…For instance, bilinear models were considered to handle complex scenarios such as multilayered scenes [2,3], by introducing additional interaction terms in the linear model. An unmixing algorithm based on a manifold learning process was investigated in [4], under the assumption that hyperspectral data may be embedded into a low-dimensional manifold. Kernel methods aim to avoid high computational complexity by using more simple physics-inspired models.…”

Section: Introductionmentioning

confidence: 99%

Estimating abundance fractions of materials in hyperspectral images by fitting a post-nonlinear mixing model

Chen

Richard

Honeiné

2013

2013 5th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS)

View full text Add to dashboard Cite

Within the area of hyperspectral data processing, nonlinear unmixing techniques have emerged as promising alternatives for overcoming the limitations of linear methods. In this paper, we consider the class of post-nonlinear mixing models of the partially linear form. More precisely, these composite models consist of a linear mixing part and a nonlinear fluctuation term defined in a reproducing kernel Hilbert space, both terms being parameterized by the endmember spectral signatures and their respective abundances. These models consider that the reproducing kernel may also depend advantageously on the fractional abundances. An iterative algorithm is then derived to jointly estimate the fractional abundances and to infer the nonlinear functional term.Index Terms-Nonlinear unmixing, post-nonlinear mixing model, kernel methods, hyperspectral data processing

show abstract

Exploiting manifold geometry in hyperspectral imagery

Cited by 398 publications

References 33 publications

Active Learning via Multi-View and Local Proximity Co-Regularization for Hyperspectral Image Classification

Active Learning via Multi-View and Local Proximity Co-Regularization for Hyperspectral Image Classification

Subspace Feature Analysis of Local Manifold Learning for Hyperspectral Remote Sensing Images Classification

Estimating abundance fractions of materials in hyperspectral images by fitting a post-nonlinear mixing model

Contact Info

Product

Resources

About