- Количество слайдов: 29
Spectral Angel Mapper (SAM) Algorithm for Landuse Mapping Partha Pratim Ghosh Product Specialist, ESRI India Dr. Deb Jyoti Pal Vice President, ESRI India Dr. Pabitra Banik Professor, Indian Statistical Institute, Kolkata Dr. Nilanchal Patel Professor & Head, Birla Institute of Technology, Ranchi
Agenda §Hypothesis §Research Question §Advantages of Spectral Angel Mapper (SAM) §Minimum Noise Fraction (MNF) §Pixel Purity Index (PPI) §n-Dimensional Visualizer (n-D) §Endmember Collection §Classification §Result §Conclusion
Hypothesis ØLand use and land management practices have a major impact on natural resources including water, soil, nutrients, plants and animals. Accurate Land use information must be develop for accurate policy making. ØDigital Image classification is one of the well accepted method to extract Land use information system and many limitation like mixed pixel and noise issues has been observed in the conventional pixel based classification techniques.
Research Question 1. Can we use spectra based classification techniques for multispectral image to develop an accurate land use information system? 2. Can we overcome the issues related to mixed pixel specifically observed in case of different type of vegetation? 3. Can we identify crops using spectra?
Study Area Eastern Part of Eastern Plateau Area (Purulia District of West Bengal) Landsat ETM+ Image Purulia District, West Bengal India Image
Landcover Land cover is the physical material at the surface of the earth, which naturally cover the earth surface. e. g. grass, asphalt, trees, bare ground, water, etc. .
Landuse “The arrangements, activities and inputs people undertake in a certain land cover type to produce, change or maintain it is called land use" (FAO, 1997 a; FAO/UNEP, 1999). Land use involves the management and modification of natural environment or wilderness into built environment such as fields, pastures, and settlements.
LULC Mapping Techniques • Survey • Pixel based Satellite Image Classification using Remote Sensing Software ØSupervised üParallelpiped üMinimum Distance üMaximum Likelihood üMahalanobis Distance ØUnsupervised üIso. Data üK-Means • Spectra based üSpectral Angel Mapper (SAM) üSpectral Information Divergence (SID)
Limitation of Pixel Based Classification Each pixel in the image is compared to the training site signatures identified by the analyst and labeled as the class it most closely "resembles" digitally. Class 1 Vegetation Urban Water Class 2 Urban Vegetation Class 3 Water
Spectral Angel Mapper (SAM) ØSAM is an automated method for comparing image spectra to individual spectra or to a spectral library (Boardman, unpublished data; CSES, 1992; Kruse et al. , 1993 a). Image Spectra Laboratory Spectra Image Spectra & Laboratory Spectra are matching ØThe algorithm determines the similarity between two spectra by calculating the spectral angle between them, treating them as vectors in n-D space, where n is the number of bands.
Spectral Angel Mapper (SAM) In a two-dimensional feature space defined by bands x and y, two spectral signatures that represent two different surface objects can be represented as vectors v 1, and v 2. Then the spectral distance (Euclidean distance) is the length of the line segment d connecting the end points of the two vectors v 1 and v 2. The spectral angle is the angle between the two vectors v 1 , and v 2 : i. e. , θ (v 1, v 2)=Cos -1 v 1 Tv 2 v 1 v 2
Spectral Angel Mapper (SAM) ØIf we linearly scale the length of vectors v 1 and v 2, by distance r, the spectral distance will be scaled by r. ØOn the other hand the cosine of the angle θ between the two vectors v 1 and v 2, remains the same. ØBecause of this invariant nature of the cosine of the angle θ to the linearly scaled variations, it becomes sensitive to the shape of the spectral patterns. Sohn et al. (1999)
Spectral Angel Mapper (SAM) ØSmall spectral angel (Cos θ) between the two spectrums indicate high similarity and high angles indicate low similarity. ØThe spectra of the same type of surface objects are approximately linearly scaled variations of one another due to the atmospheric and topographic variations. So the actual vectors in feature space will fall slightly above or below the linearly scaled vectors. But the changes in the cosine of the angle θ caused by these variations remain very small (Sohn et al. , 1999).
Method Atmospheric Correction of Image Minimum Noise Fraction (MNF) Pixel Purity Index (PPI) n-Dimensional Visualizer (n-D) Endmember Collection Creation of Unidentified Spectral Library Classification using SAM Class Identification
Minimum Noise Fraction (MNF) MNF transform is used to segregate noise in the data, and to reduce the computational requirements for subsequent processing (Boardman and Kruse, 1994). The MNF transform as modified from Green et al. (1988) and used in ENVI
Pixel Purity Index (PPI) • The Pixel Purity Index (PPI) is a means of finding the most “spectrally pure, ” or extreme, pixels in multispectral and hyperspectral images (Boardman et al. , 1995). • The Pixel Purity Index records the total number of times each pixel is marked as extreme. A "Pixel Purity Image" is created in which the DN of each pixel corresponds to the number of times that pixel was recorded as extreme.
n-Dimensional Visualizer ØSpectra can be thought of as points in an n -dimensional scatter plot, where n is the number of bands. ØThe n-D Visualizer help to visualize the shape of a data cloud that results from plotting image data in spectral space (with image bands as plot axes). ØWe typically used the n-D Visualizer with spatially subsetted Minimum Noise Fraction (MNF) data that use only the purest pixels determined from the Pixel Purity Index (PPI). .
n-Dimensional Visualizer ØRotating n-D Visualizer interactively we can select groups of pixels in classes. Selected classes can be exported to us in the classification. Øn-D Visualizer can be used to check the separability of the classes when the regions of interest (ROIs) as input into supervised. ØThe n-D Visualizer is an interactive tool to use for selecting the endmembers in n-D space.
Endmember Collection Endmembers are spectra that are chosen to represent pure surface materials in a spectral image. Endmembers that represent radiance or reflectance spectra must satisfy a positivity constraint (containing no values less than zero).
SAM Classification Use the Endmember in Spectral Angel Mapper Algorithm
Class Identification Surveyed villages and markets in Purulia District Legend: Surveyed villages District market Major markets Minor markets Railway Road Block Boundary Data source: ISI, Calcutta, India
Class Identification Landuse Ecosystem Pattern of Kashipur Block
Conclusion 1. Even if Landsat ETM+ is a medium spatial resolution and that sub-pixel contamination cover material is evident while selecting endmembers, it has given good results in SAM. 2. The classification map generated with SAM for Landsat ETM+ show that this method could effectively be used for landuse mapping. 3. With the help of MNF, PPI & n-D Visualizer the mixed pixel issue can be addressed up to certain level
References 1. Natural Resource Inventory of Luppi Village, Eastern Plateau of India: Implications for Sustainable Agricultural Development by P. Banik, Ashim Midhya, Sharon Fajardo, P. Suan Kam --- Journal of Sustainable Agriculture, Volume 28, Issue 2, Page 85 – 100. 2. Poverty Reduction in the Tribal Belt of Eastern India by Christopher Edmonds, Nobuhiko Fuwa, P. Banik ----Asia Pacific Issues, No. 81, Honolulu: East-West Center, August 2006. 3. Rejuvenation of agriculture in India: Cost benefits in using EO products by V. Jayaraman, J. S. Parihar and S. K. Srivastava ---- Acta Astronautica, Volume 63, Issues 1 -4, July-August 2008, Pages 493 -502. 4. Watershed externalities, shifting cropping patterns and groundwater depletion in Indian semiarid villages: The effect of alternative water pricing policies by Bekele Shiferaw, V. Ratna Reddy and Suhas P. Wani ---- www. sciencedirect. com 5. Distributed ecohydrological modelling to evaluate irrigation system performance in Sirsa Sistrict, India II: Impact of viable water management scenarios by R. Singh, R. K. Jhorar, J. C. van Dam and R. A. Feddes ---- Journal of Hydrology, Volume 329, Issues 3 -4, 15 October 2006, Pages 714 -723 27
References 6. Patterns and ecological implications of agricultural land-use changes: a case study from central Himalaya, India by R. L. Semwal, S. Nautiyal, K. K. Sen, U. Rana, R. K. Maikhuri, K. S. Rao and K. G. Saxena ----Agriculture, Ecosystems & Environment, Volume 102, Issue 1, March 2004, Pages 81 -92. 7. National spatial crop yield simulation using GIS-based crop production model by Satya Priya, and Ryosuke Shibasaki ---- Ecological Modelling, Volume 136, Issues 2 -3, 20 January 2001, Pages 113 -129. 8. The role of GIS and remote sensing in land degradation assessment and conservation mapping: some user experiences and expectations by Godert W. J. van Lynden and Stephan Mantel ----International Journal of Applied Earth Observation and Geoinformation Volume 3, Issue 1, 2001, Pages 61 -68. 9. National spatial data infrastructure - coming together of GIS and EO in India by Mukund Rao, Amitabha Pandey, A. K. Ahuja, V. S. Ramamurthy and K Kasturirangan ---- Acta Astronautica, Volume 51, Number 1, July 2002 , pp. 527 -535(9) 10. Integration of remotely sensed and model data to provide the spatial information basis for sustainable landuse by R. Backhaus and G. Braun ---- Acta Astronautica, Volume 42, Issue 9, May 1998, Pages 541 -546. 11. Spatial-Temporal Pattern and Driving Forces of Land Use Changes in Xiamen by Bin QUAN, , Jian-Fei CHEN, Hong-Lie QIU, M. J. M. RÖMKENS, Xiao-Qi YANG, Shi-Feng JIANG and Bi-Cheng LI ----Pedosphere, Volume 16, Issue 4, August 2006, Pages 477 -488. 28