China
Africa
Explore chapters and articles related to this topic
Ecological Characterization of Vegetation Using Multisensor Remote Sensing in the Solar Reflective Spectrum
Published in Prasad S. Thenkabail, Land Resources Monitoring, Modeling, and Mapping with Remote Sensing, 2015
Conghe Song, Jing Ming Chen, Taehee Hwang, Alemu Gonsamo, Holly Croft, Quanfa Zhang, Matthew Dannenberg, Yulong Zhang, Christopher Hakkenberg, Juxiang Li
e major challenge in the use of SMA is the selection of appropriate endmembers and their spectral signatures. Endmembers can be derived either directly from remotely sensed imagery (image endmembers) (e.g., DeFries et al. 1999; Song 2004) or from ¤eld or laboratory measurements (e.g., Adams et al. 1995; Roberts et al. 1998). e number of endmembers that can be used is limited by the dimensionality of the remotely sensed image data. In the case of Landsat imagery, for example, SMA techniques are generally limited to 3-4 endmembers. In many complex landscapes, 3-5 endmembers may be insu¶cient to represent the spectral and spatial variability within an image. A variety of techniques exist to account for endmember variability (reviewed in Somers et al. 2011), including the multiple endmember SMA technique (Roberts et al. 1998), in which endmember models are selected separately for each pixel in the image from a large library of spectral endmembers to construct numerous candidate models, from which the “best” candidate model is selected for each pixel to perform SMA. Somers et al. (2011) suggest that these types of iterative endmember selection approaches can provide a more e¦ective representation of endmember variability than simple SMA approaches (in which endmember signatures are assumed constant across the entire image). Song (2005) developed the Bayesian spectral mixture analysis (BSMA) to account for endmember signature variation when estimating fc in a pixel. e endmember spectral signature in BSMA is represented by a probability mass function instead of a constant. Deng and Wu (2013) further developed an algorithm that adaptively generates endmember spectral signatures over space to account for endmember spectral signature variations.
Image Unmixing and Segmentation
Published in Gerhard X. Ritter, Gonzalo Urcid, Introduction to Lattice Algebra, 2021
Gerhard X. Ritter, Gonzalo Urcid
For a hyperspectral image of size k=p×q pixels acquired over n spectral bands, the computational effort required for pixel spectra extraction (task 1) is linear in k since n≪k. The overall computational complexity of the WM based technique (tasks 2,3,4) is n2(k+3), which for values of n of a few hundreds is in the order of few minutes (see Table 7.4). Endmember determination (task 5) relies on a subset of 2⌊n+1⌋ candidate extremal vectors generated after task 4 is completed and, although interactive in nature, m∝⌊n⌋ “final” endmembers can be selected in a lapse of minutes. On the other hand, the NNLS method needs about nm3 arithmetical operations to find a unique set of abundance coefficients for each pixel spectra. Since m represents the number of final endmembers and m≪k, it turns out that for hyperspectral image segmentation the computational complexity of the non-negative least square method (task 6) is nm3 per pixel. Finally, general resource classification (task 7) may also be accomplished within minutes whenever a working knowledge in spectral identification or prior experience in hyperspectral image analysis is available for the problem at hand [129].
Petrology, geochemistry and a probable late Cambrian age for harzburgites of the Coolac Serpentinite, New South Wales, Australia
Published in Australian Journal of Earth Sciences, 2018
Minerals were analysed for major elements including Ni, Cr, Zn and V using a Cameca SX 100 electron microprobe at Macquarie University. Analytical conditions were optimised for a standard silicate (olivine and pyroxene) and non-silicate (Cr-spinel) run using a 15 keV accelerating voltage and a 20 nA focussed electron beam. Routine analyses for the silicates were obtained by counting 20 s at peak and 20 s on background. For the spinels, elemental counting times were the same except for Ti, which had 60 s on peak count time. Natural and synthetic materials were run as standards and were well within the ±10% relative margin for each element. Instrumental drift was negligible. Total iron is reported as FeO. Fe2+ and Fe3+ were calculated for olivine and pyroxene by charge balance and for spinel by assuming perfect stoichiometry and endmember proportions.
Cretaceous tungsten-tin mineralisation in the Tin Range, Stewart Island, New Zealand
Published in New Zealand Journal of Geology and Geophysics, 2022
Hamish C. Lilley, James M. Scott, Josh J. Schwartz, Rose E. Turnbull, Andy J. Tulloch
The Tin Range granodiorite includes both equigranular (3–5 mm) and megacrystic areas (1–3 cm) with a dominant mineralogy of K-feldspar, plagioclase, quartz, biotite, white mica, garnet and apatite (Figure 4D,E) (Henley and Higgins 1977). The megacrystic parts include phenocrysts of either euhedral to subhedral almandine-spessartine garnet (semi-quantitatively: Fe80Mn17Ca3) ranging from ∼0.5 to 1 cm in size or 1 to 3 cm megacrystic aggregates of subhedral to anhedral K-feldspar and plagioclase. The feldspars are generally near endmember orthoclase and albite, but perthitic feldspar with ∼10 µm lamellae is also present in minor quantities.
Mapping of mineral resources and lithological units: a review of remote sensing techniques
Published in International Journal of Image and Data Fusion, 2019
Rejith Rajan Girija, Sundararajan Mayappan
The ASA is a six-fold spectral analysis technique applied to hyperspectral as well as multispectral RS data for identifying mineral resources. It includes six continuous processes of (1) reflectance calibration; (2) minimum noise transformation; (3) pixel purity index (PPI); (4) n-d visualise; (5) identification of end member spectra using spectral techniques like SFF, SAM and binary encoding (BE); (6) spectral mapping of mineral deposits using algorithms like SAM and MTMF (Pal et al. 2011). MNF transformation is a doubly cascaded PCA technique used for linearly transforming the reflectance-converted data in order to reduce its spectral dimensionality and noise (Green et al. 1988). It calculates the noise statistics of the input data in terms of eigen values and these values were used for selecting the lower MNF bands containing the maximum spectral information. PPI separates the most spectrally pure pixels from these MNF bands and generates a PPI image in which the values of each pixel correspond to the number of times the pixel was recorded as extreme. Thus the PPI reduces the number of pixel in the input data which leads to the selection of spectrally unique target members or endmembers. The purest pixels were located as pixel clouds at the corner of the n-D scatter plot of the n-D visualizer. These pixel clouds can be rotated and visualised in different directions and angles which help to identify and isolate target endmembers as group of unique corner points cloud. The points cloud was exported as ROI in the image and end member spectra was extracted (Kruse et al. 2003, Research Systems Inc. 2003). The extracted endmembers spectra were compared with existing mineral libraries of United States Geological Survey, John Hopkins University and Jet Propulsion Laboratory. The matching of endmember spectra with various mineral types exists in the above libraries were done using the techniques like SFF, SAM and BE which produce a score between 0 and 1, where the value 1 shows a perfect match with the mineral type exist in the libraries. Finally these verified target endmembers are mapped using spectral classification techniques. Flowchart showing different steps of ASA technique is shown in Figure 2.