Sparse Dictionary Learning for Blind Hyperspectral Unmixing

Liu, Yang; Guo, Yi; Li, Feng; Lei, Xin; Huang, Puming

doi:10.1109/lgrs.2018.2878036

Cited by 7 publications

(6 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This algorithm is also known as the fast non-negative orthogonal matching pursuit (FNNOMP). To learn EMs with the C-SCD model, two fundamental changes were necessary [16]. The p norm constraint from atoms is removed because real EMs often do not hold such constraint.…”

Section: Prior Work In Dictionary Learning (Dl)mentioning

confidence: 99%

“…There are two main approaches towards the solution of Equations ( 1 ) and ( 2 ): one is to find the purest element in the scene through searching for the convex cone of the spectral data (see [ 6 , 7 ] for an overview), and the other is through the optimisation of the sparsity of the abundance “a” [ 16 ] or using greedy algorithm for learning the dictionary

. Classical search methods that exploit the convex distribution property of data such as the vertex component analysis (VCA) [ 17 ], more recent algorithms like the minimum volume simplex analysis (MVSA) [ 18 ] and the Collaborative Nonnegative Matrix Factorization (CoNMF) [ 19 ] etc., have provided good solutions especially when relatively pure pixels are present in the scene.…”

Section: Prior Work In Dictionary Learning (Dl)mentioning

confidence: 99%

“…To learn EMs with the C-SCD model, two fundamental changes were necessary [ 16 ]. The

norm constraint from atoms is removed because real EMs often do not hold such constraint.…”

Section: Prior Work In Dictionary Learning (Dl)mentioning

confidence: 99%

See 2 more Smart Citations

Endmember Learning with K-Means through SCD Model in Hyperspectral Scene Reconstructions

Chatterjee

Yuen

2019

J. Imaging

View full text Add to dashboard Cite

This paper proposes a simple yet effective method for improving the efficiency of sparse coding dictionary learning (DL) with an implication of enhancing the ultimate usefulness of compressive sensing (CS) technology for practical applications, such as in hyperspectral imaging (HSI) scene reconstruction. CS is the technique which allows sparse signals to be decomposed into a sparse representation “a” of a dictionary D u . The goodness of the learnt dictionary has direct impacts on the quality of the end results, e.g., in the HSI scene reconstructions. This paper proposes the construction of a concise and comprehensive dictionary by using the cluster centres of the input dataset, and then a greedy approach is adopted to learn all elements within this dictionary. The proposed method consists of an unsupervised clustering algorithm (K-Means), and it is then coupled with an advanced sparse coding dictionary (SCD) method such as the basis pursuit algorithm (orthogonal matching pursuit, OMP) for the dictionary learning. The effectiveness of the proposed K-Means Sparse Coding Dictionary (KMSCD) is illustrated through the reconstructions of several publicly available HSI scenes. The results have shown that the proposed KMSCD achieves ~40% greater accuracy, 5 times faster convergence and is twice as robust as that of the classic Spare Coding Dictionary (C-SCD) method that adopts random sampling of data for the dictionary learning. Over the five data sets that have been employed in this study, it is seen that the proposed KMSCD is capable of reconstructing these scenes with mean accuracies of approximately 20–500% better than all competing algorithms adopted in this work. Furthermore, the reconstruction efficiency of trace materials in the scene has been assessed: it is shown that the KMSCD is capable of recovering ~12% better than that of the C-SCD. These results suggest that the proposed DL using a simple clustering method for the construction of the dictionary has been shown to enhance the scene reconstruction substantially. When the proposed KMSCD is incorporated with the Fast non-negative orthogonal matching pursuit (FNNOMP) to constrain the maximum number of materials to coexist in a pixel to four, experiments have shown that it achieves approximately ten times better than that constrained by using the widely employed TMM algorithm. This may suggest that the proposed DL method using KMSCD and together with the FNNOMP will be more suitable to be the material allocation module of HSI scene simulators like the CameoSim package.

show abstract

Section: Prior Work In Dictionary Learning (Dl)mentioning

confidence: 99%

Section: Prior Work In Dictionary Learning (Dl)mentioning

confidence: 99%

See 1 more Smart Citation

Endmember Learning with K-Means through SCD Model in Hyperspectral Scene Reconstructions

Chatterjee

Yuen

2019

J. Imaging

View full text Add to dashboard Cite

show abstract

“…For instance, the high dimensionality of HSI data can be leveraged using deep feature extraction techniques (Chen et al, 2016;Rasti et al, 2020) that transform raw data in a hierarchical fashion to a lower dimensional data representation composed of new variables that are more discriminant, abstract and robust. Challenges from spectral mixing in HSI data can be minimized using dictionary learning-based unmixing approaches (Hong et al, 2019;Liu et al, 2019) to understand the material composition of a single pixel. Even in the presence of signal noise, targets of interest can readily be detected using ensemble learning techniques (Zhao et al, 2019;Sun et al, 2020) and classified using graph convolutional neural networks (Qin et al, 2019;Hong et al, 2020a).…”

Section: Introductionmentioning

confidence: 99%

The Directly-Georeferenced Hyperspectral Point Cloud: Preserving the Integrity of Hyperspectral Imaging Data

Inamdar

Kalácska

Arroyo‐Mora

et al. 2021

Front. Remote Sens.

View full text Add to dashboard Cite

The raster data model has been the standard format for hyperspectral imaging (HSI) over the last four decades. Unfortunately, it misrepresents HSI data because pixels are not natively square nor uniformly distributed across imaged scenes. To generate end products as rasters with square pixels while preserving spectral data integrity, the nearest neighbor resampling methodology is typically applied. This process compromises spatial data integrity as the pixels from the original HSI data are shifted, duplicated and eliminated so that HSI data can conform to the raster data model structure. Our study presents a novel hyperspectral point cloud data representation that preserves the spatial-spectral integrity of HSI data more effectively than conventional square pixel rasters. This Directly-Georeferenced Hyperspectral Point Cloud (DHPC) is generated through a data fusion workflow that can be readily implemented into existing processing workflows used by HSI data providers. The effectiveness of the DHPC over conventional square pixel rasters is shown with four HSI datasets. These datasets were collected at three different sites with two different sensors that captured the spectral information from each site at various spatial resolutions (ranging from ∼1.5 cm to 2.6 m). The DHPC was assessed based on three data quality metrics (i.e., pixel loss, pixel duplication and pixel shifting), data storage requirements and various HSI applications. All of the studied raster data products were characterized by either substantial pixel loss (∼50–75%) or pixel duplication (∼35–75%), depending on the resolution of the resampling grid used in the nearest neighbor methodology. Pixel shifting in the raster end products ranged from 0.33 to 1.95 pixels. The DHPC was characterized by zero pixel loss, pixel duplication and pixel shifting. Despite containing additional surface elevation data, the DHPC was up to 13 times smaller in file size than the corresponding rasters. Furthermore, the DHPC consistently outperformed the rasters in all of the tested applications which included classification, spectra geo-location and target detection. Based on the findings from this work, the developed DHPC data representation has the potential to push the limits of HSI data distribution, analysis and application.

show abstract

“…Now, dictionary learning-based endmember extraction approaches have become a popular research topic. A spectral unmixing method based on sparse dictionary learning is proposed in [19], which obtains good results in the observed image reconstruction. However, this approach requires quite a high signal to noise ratio (SNR).…”

Section: Introductionmentioning

confidence: 99%

A Novel Hyperspectral Endmember Extraction Algorithm Based on Online Robust Dictionary Learning

Xiaorui

2019

Remote Sensing

View full text Add to dashboard Cite

Due to the sparsity of hyperspectral images, the dictionary learning framework has been applied in hyperspectral endmember extraction. However, current endmember extraction methods based on dictionary learning are not robust enough in noisy environments. To solve this problem, this paper proposes a novel endmember extraction approach based on online robust dictionary learning, termed EEORDL. Because of the large scale of the hyperspectral image (HSI) data, an online scheme is introduced to reduce the computational time of dictionary learning. In the proposed algorithm, a new form of the objective function is introduced into the dictionary learning process to improve the robustness for noisy HSI data. The experimental results, conducted with both synthetic and real-world hyperspectral datasets, illustrate that the proposed EEORDL outperforms the state-of-the-art approaches under different signal-to-noise ratio (SNR) conditions, especially for high-level noise.

show abstract

Sparse Dictionary Learning for Blind Hyperspectral Unmixing

Cited by 7 publications

References 12 publications

Endmember Learning with K-Means through SCD Model in Hyperspectral Scene Reconstructions

Endmember Learning with K-Means through SCD Model in Hyperspectral Scene Reconstructions

The Directly-Georeferenced Hyperspectral Point Cloud: Preserving the Integrity of Hyperspectral Imaging Data

A Novel Hyperspectral Endmember Extraction Algorithm Based on Online Robust Dictionary Learning

Contact Info

Product

Resources

About