In spectral unmixing theory, data reduction techniques play an important role as hyperspectral imagery contains an immense amount of data, posing many challenging problems such as data storage, computational efficiency, and the so called “curse of dimensionality”. Feature extraction and feature selection are the two main approaches for dimensionality reduction. Feature extraction techniques are used for reducing the dimensionality of the hyperspectral data by applying transforms on hyperspectral data. Feature selection techniques retain the physical meaning of the data by selecting a set of bands from the input hyperspectral dataset, which mainly contain the information needed for spectral unmixing. Although feature selection techniques are well-known for their dimensionality reduction potentials they are rarely used in the unmixing process. The majority of the existing state-of-the-art dimensionality reduction methods set criteria to the spectral information, which is derived by the whole wavelength, in order to define the optimum spectral subspace. These criteria are not associated with any particular application but with the data statistics, such as correlation and entropy values. However, each application is associated with specific land c over materials, whose spectral characteristics present variations in specific wavelengths. In forestry for example, many applications focus on tree leaves, in which specific pigments such as chlorophyll, xanthophyll, etc. determine the wavelengths where tree species, diseases, etc., can be detected. For such applications, when the unmixing process is applied, the tree species, diseases, etc., are considered as the endmembers of interest. This paper focuses on investigating the effects of band selection on the endmember extraction by exploiting the information of the vegetation absorbance spectral zones. More precisely, it is explored whether endmember extraction can be optimized when specific sets of initial bands related to leaf spectral characteristics are selected. Experiments comprise application of well-known signal subspace estimation and endmember extraction methods on a hyperspectral imagery that presents a forest area. Evaluation of the extracted endmembers showed that more forest species can be extracted as endmembers using selected bands.
In the hyperspectral theory, data reduction techniques play an important role in the classification processing as hyperspectral imagery contains an immense amount of data posing many challenging problems such as data storage, computational efficiency, and the curse of dimensionality. Hyperspectral band selection technique is a well-known dimensionality reduction approach which retains the physical meaning of the data. It selects a set of bands from the input hyperspectral dataset which comprises the information needed for subsequent hyperspectral image spectroscopy. The majority of the existing state-of-the-art dimensionality reduction methods set criteria to the spectral information which is derived by the whole wavelength in order to define the optimum spectral subspace. These criteria are not associated with the particular classification task but with the data statistics, such as correlation and entropy values. However, each spectral signature of a particular material has spectral characteristics which contribute to distinguish it from other spectral signatures at specific sequential wavelengths. This paper focuses on investigating the effects of band selection on the classification by exploiting the information of sequential bands. More precisely, it is explored 1) whether classification can be optimized when a different set of initial bands is selected per category; 2) whether there is an optimum subset of sequential bands which lead to more accurate classification results. Experiments comprise application of a well-known classification method, the support vector machine (SVM), on real hyperspectral dataset using all the possible subsets of p sequential bands, where p is equal to the dimensionality of the signal subspace. Evaluation of the classification accuracy leads to remarkable conclusions.
Oil spill events are a crucial environmental issue. Detection of oil spills is important for both oil exploration and
environmental protection. In this paper, investigation of hyperspectral remote sensing is performed for the detection of
oil spills and the discrimination of different oil types. Spectral signatures of different oil types are very useful, since they
may serve as endmembers in unmixing and classification models. Towards this direction, an oil spectral library, resulting
from spectral measurements of artificial oil spills as well as of look-alikes in marine environment was compiled. Samples
of four different oil types were used; two crude oils, one marine residual fuel oil, and one light petroleum product. Lookalikes
comprise sea water, river discharges, shallow water and water with algae. Spectral measurements were acquired
with spectro-radiometer GER1500. Moreover, oil and look-alikes spectral signatures have been examined whether they
can be served as endmembers. This was accomplished by testifying their linear independence. After that, synthetic
hyperspectral images based on the relevant oil spectral library were created. Several simplex-based endmember
algorithms such as sequential maximum angle convex cone (SMACC), vertex component analysis (VCA), n-finder
algorithm (N-FINDR), and automatic target generation process (ATGP) were applied on the synthetic images in order to
evaluate their effectiveness for detecting oil spill events occurred from different oil types. Results showed that different
types of oil spills with various thicknesses can be extracted as endmembers.