The applications of hyperspectral infrared imagery in the different fields of research are significant and growing. It is mainly used in remote sensing for target detection, vegetation detection, urban area categorization, astronomy and geological applications. The geological applications of this technology mainly consist in mineral identification using in airborne or satellite imagery. We address a quantitative and qualitative assessment of mineral identification in the laboratory conditions. We strive to identify nine different mineral grains (Biotite, Diopside, Epidote, Goethite, Kyanite, Scheelite, Smithsonite, Tourmaline, Quartz). A hyperspectral camera in the Long Wave Infrared (LWIR, 7.7-11.8 ) with a LW-macro lens providing a spatial resolution of 100 <i>μm</i>, an infragold plate, and a heating source are the instruments used in the experiment. The proposed algorithm clusters all the pixel-spectra in different categories. Then the best representatives of each cluster are chosen and compared with the ASTER spectral library of JPL/NASA through spectral comparison techniques, such as Spectral angle mapper (SAM) and Normalized Cross Correlation (NCC). The results of the algorithm indicate significant computational efficiency (more than 20 times faster) as compared to previous algorithms and have shown a promising performance for mineral identification.
Hyperspectral imaging (HSI) in the long-wave infrared spectrum (LWIR) provides spectral and spatial information concerning the emissivity of the surface of materials, which can be used for mineral identification. For this, an endmember, which is the purest form of a mineral, is used as reference. All pure minerals have specific spectral profiles in the electromagnetic wavelength, which can be thought of as the mineral’s fingerprint. The main goal of this paper is the identification of minerals by LWIR hyperspectral imaging using a machine learning scheme. The information of hyperspectral imaging has been recorded from the energy emitted from the mineral’s surface. Solar energy is the source of energy in remote sensing, while a heating element is the energy source employed in laboratory experiments. Our work contains three main steps where the first step involves obtaining the spectral signatures of pure (single) minerals with a hyperspectral camera, in the long-wave infrared (7.7 to 11.8 μm), which measures the emitted radiance from the minerals’ surface. The second step concerns feature extraction by applying the continuous wavelet transform (CWT) and finally we use support vector machine classifier with radial basis functions (SVM-RBF) for classification/identification of minerals. The overall accuracy of classification in our work is 90.23± 2.66%. In conclusion, based on CWT’s ability to capture the information of signals can be used as a good marker for classification and identification the minerals substance.
There are numerous image fusion techniques such as intensity-hue-saturation (IHS) transform and principal component analysis (PCA). These methods are offering promising performance but the drawback with them is that they are not necessarily optimal in newer applications such as Ikonos and QuickBird. Color distortion is of vital importance in fusion image processing. The main result of this paper is the development of a fast HPM-enhanced version of the IHS method for application in fusion image processing in high-resolution satellite images. Combining these two methods makes it possible to benefit from the advantages of both methods. To evaluate the HPM-enhanced version of IHS method we used QuickBird data. The HPM-enhanced version of IHS and HPM-enhanced IHS are used interchangeably. The simulation results of this method show that it is capable of providing a significant improvement in preserving spectral and spatial information.
Hyperspectral imaging has been considerably developed during the recent decades. The application of hyperspectral imagery and infrared thermography, particularly for the automatic identification of minerals from satellite images, has been the subject of several interesting researches. In this study, a method is presented for the automated identification of the mineral grains typically used from satellite imagery and adapted for analyzing collected sample grains in a laboratory environment. For this, an approach involving Sparse Principle Components Analysis (SPCA) based on spectral abundance mapping techniques (i.e. SAM, SID, NormXCorr) is proposed for extraction of the representative spectral features. It develops an approximation of endmember as a reference spectrum process through the highest sparse principle component of the pure mineral grains. Subsequently, the features categorized by kernel Extreme Learning Machine (Kernel- ELM) classify and identify the mineral grains in a supervised manner. Classification is conducted in the binary scenario and the results indicate the dependency to the training spectra.
The proposed approach addresses the problem of retrieving the emissivity of hyperspectral data in the spectroscopic imageries from indoor experiments. This methodology was tested on experimental data that have been recorded with hyperspectral images working in visible/near infrared and long-wave infrared bands. The proposed technique provides a framework for computing down-welling spectral radiance applying non-negative matrix factorization (NMF) analysis. It provides the necessary means for the non-uniform correction of active thermographical experiments. The obtained results indicate promising accuracy. In addition, the application of the proposed technique is not limited to non-uniform heating spectroscopy but to uniform spectroscopy as well.