Upload
psychosight
View
11
Download
0
Tags:
Embed Size (px)
DESCRIPTION
spring
Citation preview
GIS practical using SPRING
Michael Gould
Lubia Vinhas
Remote Sensing Images
GeoTiffTagged Image File Format is a file format for storing images within a single file. It contains
header tags (size, definition, image-data arrangement, applied image compression) defining
the image's geometry
Remote Sensing Images
GeoTiffTIFF based interchange format for georeferenced
raster imagery. The potential additional information includes projections, coordinate systems, ellipsoids, datums, and everything else necessary to establish
the exact spatial reference for the file.
Remote Sensing Images
GeoTiffTIFF based interchange format for georeferenced
raster imagery. The potential additional information includes projections, coordinate systems, ellipsoids, datums, and everything else necessary to establish
the exact spatial reference for the file.
SPRING – P1
- Create a new SPRING database (remember: it is just a repository!)
- Create a partition where you are going to store your images, and call it CBERS_IMG.
SPRING
- Import the files *Belem_1.tif, Belem_2.tif, Belem_3.tif and Belem_4.tif to your database:- When you import the first file, a new Project will be
created with the bounding box and projection defined in the geotiff
- This images are images from the CCD instrument of the CBERS satellite in 10/08/2008
- http://www.cbers.inpe.br/?hl=en&content=cameras1e2e2b for more information about the instrument
*Belém is the capital of Pará state in the north of Brazil
Remember the reflectance curve
CBERS CCD Bands
Band 1 = 0,45 - 0,52 m (blue)
Band 2 = 0,52 - 0,59 m (green)
Band 3 = 0,63 - 0,69 m (red)
Band 4 = 0,77 - 0,89 m (near infrared)
Band 5 = 0,51 - 0,73 m (pancromatic)
SPRING
- Display each band separately in monocromatic mode
- Find the best (visually) color composition using this bands
Segmentation
Image is partitioned into meaningful regions whose points have nearly the same properties,e.g., grey levels, mean values or textural properties
Segmentation methods
1) Region-based: dividing up the images into a number of homogeneous regions, each having a unique label
2) Edge detection: determining boundaries between homogeneous regions of different properties
3) Combination of region-based and edge detection
The selection of any of these segmentation approaches
greatly depends on the type of data being analyzed and
on the application area
Region Growing
1. Segment the entire image into pattern cells (1 or more pixels).
2. Each pattern cell is compared with its neighboring cells to
determine if they are similar, using a similarity measure. If they
are similar, merge the cells to form a fragment and update the
property used in the comparison.
3. Continue growing the fragment by examing all of its neighbors
until no joinable regions remain. Label the fragment as a
completed region.
4. Move to the next uncompleted cell, and repeat these steps until
all cells are labeled.
Region Growing in SPRING
ParametersSimilarity: Treshold value below which two regions are
considered similar at a given step, considering the similarity as the Euclidean distance between the mean vector of two regions
Area: Minimun number of pixels allowed in an area
OutputNew layer with the labeled regions
Run two segmentation experiments on the images that you have. Compare the results.
Edge Detection
• Edges are boundaries in the image. Are areas with strong intensity contrasts.
Sobel operator
Aproximates the gradient by convolution masks
-1 0 1
-2 0 2
-1 0 1
1 2 1
0 0 0
-1 -2 -1
Gx Gy
G = Gx + Gy
Watershed
The gradient image is considered as a topographic surface
Flood this surface from its minima and, preventing the merging of the waters coming from different sources, two different sets: the catchment basins and the watershed lines
Region Growing in SPRING
ParametersInitial Level: the initial value from where the
flooading starts
Output
New layer with the labeled regions
Run two segmentation experiments on the images that you have. Compare the results.
Classification
The extraction of distincts classes or themes related to land use and land cover categories
Urban
Water
Bare soil
Classifiers
Pixel-By-Pixel: consider each pixel as an object that has to be classified
Region: consider each a homogeneous region (previously delimitaded) as an object to classified
Supervised: samples of known targets are used to describe the spectral signature of the target, and then extrapolated to other areas of unknown targets
Unsupervised: based solely on the image statistics without availability of training samples or a-priori knowledge of the area
Pixel-by-pixel, unsupervised
K-MeansMinimize the variability within each clusterK: number of clusters (themes) that have to be defined.
It is defines a prioryInteractive algorithm:
1. Randomly assign the posicion of the centroid of each cluster
2. Assign each pixel to the nearest cluster3. Recalculate the centroid as the baricenter of the pixels in
the cluster4. Repeat steps 2 to 3 until there is no more modification or
such stop criteria is reached
Region, unsupervised
Each region calculates: mean, variance and area
Regions are ordered decreasing according to its area
A similarity value and a similarity distance is choosen
Assume that the first region defines a theme
For the next region:If it is similar to any of the founded themes assign it to that theme
If not define it as a new theme
Repeat the process until every region has been assigned to a theme
Reclassify again, after recalculating the parameters from the regions generated in the previous steps
Supervised classification
The user indicates which classes should be present in the image and choose samples of each class
Sampling:Samples should be homogeneous and should
represent the entire variability of the classNumber of samples, in general, is >> number of
classesSample analysis should be carried out before the
classification
Supervised classification
Different algorithms, same principle: samples characterize statistically the themeMinimun distance: each theme is characterized by
its mean vector. Pixels are assigned to closest theme according to a Euclidean distance
Maximun Likelihood: assume that themes have a Normal Distribution, that can be descrita by the mean and a covariance matrix calculated from the samples. Each is assigned to the theme that has the maximun probability of contain the pixel
Unsupervised Classification in SPRING
1. If regiona) Execute a segmentation
2. Create a context file3. Define bands to be used in the classification4. If supervised
a) Select samplesb) Analize the samples
5. Run the classification6. Execute post-classification procedures7. Execute mapping to classes
To be continued...