SAR Flashcards
SAR is what type of looking? Normal incidence angles?
- Side-looking
- Normally 20 - 50 degrees
How does SAR achieve spatial resolutions of less than 10m?
- Aperture synthesis
- Forward motion of craft and sensor’s ability to store target location = larger effective aperture length, or synthetic aperture
- Antenna ‘smears’ itself across Earth surface, remembers targets = increased spatial res
SAR has what type of measurements?
- Echo (amplitude)
- Phase (timing)
In pixel form, the core SAR image product is called what? Contains how many data values for every channel?
- Single-Look Complex (SLC)
- 2 data values for every channel, Amplitude and Phase
What does a channel refer to in SAR?
- Each transmit and receive polarization combo that the SAR is capable of acquiring
- eg HH and HV
Phase measurements
- Timing of phase cycles or ‘clock time’ (wavelength peak to peak?)
- Does not form intuitive image like amplitude
- Valued for more advanced applications like interferometric SAR and Polarimetric SAR
What range are SAR images collected?
- Slant range
- Must be converted to ground range to get proper image
Slant range
- Distances are measured between the antenna and the target
- Not ground distances
- Radar is essentially a ranging instrument measuring distances to objects and sigma naught in slant range
Ground Range
- Distances are measured between the platform ground track and target
- Placed in the correct position on the chosen reference frame (projection) when processed into an image
What does the slant range vs. ground range configuration lead to?
- Compression of imaged surface information in the near-range
- Leads to distortions that are large in the range direction
What are the 3 types of major distortion?
- Foreshortening
- Layover
- Shadowing
- Mainly caused by topographic variations and are called relief distortions
Where is compressional distortion most emphasized?
- Near range compression in slant range
Foreshortening
- Compression of topography in scene which are tilted toward the radar
- Found in mountainous terrain
- Radar beam reaches 2 different points in elevation at the same time, therefore appears in same position
- Ex. beam in slant range reaches base of slope before or at same time as top
Layover
- Extreme foreshortening
- Tall objects are displaced towards the radar due to their shorter distances in slant range relative to ground
- Slant beam reaches top before it reaches bottom, makes further point appear closer than positions that are nearer
- Features lean toward sensor
- Common for mountains and always occurs for buildings (90 degree angle)
Shadowing
- Areas on the ground not illuminated by radar
- Leaves dark tone in imagery
- Occurs along range direction behind tall objects
Detected format
- Name for SAR images provided in ground range corrected format
- Still subject to relief distortions, especially mountains, even after correction
- True correction would require correction at each data point for local terrain slope and elevation –> big task
What are the 2 SAR modes of operation
- Sensor modes
- Beam modes
What are the 3 main sensor modes?
- Single-beam, also called strip map
- ScanSAR
- Spotlight
Single-beam
- Wide range of incidence angles
- Image quality is balance between fine resolution and wide coverage
- Wide swath = low spatial res, narrow swath = higher res
ScanSAR
- Very wide swath at expense of spatial resolution
Spotlight
- Highest spatial resolution
- Mechanical steering of antenna allows coverage of a specific target area (squint angle)
- Spotlight allows sensor to build up high spatial res
- ‘Dwells’ on target, gets more energy, sent and received from target
Spotlight: Squint Angle
- Enables radar beam to ‘dwell’ on a footprint for a longer period of time
- Uses more pulses and increases spatial resolution w/in a footprint
- More pulses = higher spatial res
T/F: One SAR has several beam modes for each sensor mode?
True
- Eg Radarsat-2: Sensor modes spotlight, single beam, ScanSAR
Beam mode vs. nominal swath width
- Spotlight = finest resolution
- Then single beam w/ many beam modes from ultra-fine to wide
- Then ScanSAR w/ narrow and wide
Name some examples of Singlebeam modes
- Ultra-fine, fine quad, standard quad, wide standard quad, wide fine quad, fine, multi-look fine, extended high extended low, wide
Radarsat-2 Beam modes
- All available as right or left-looking
- Ultrafine, Standard beams, standard quad-pol (reduced swath width), Wide swath beams, Fine resolution beams multi-look fine resolution, fine quad pol, ScanSar (wide, narrow), Extended beams (high or low incidence)
Spotlight vs ScanSAR resolution
- Spotlight = 10x20km, .8m x 2.5m res
- ScanSar wide = 500km x 500km, 100m res
Arctic SAR application
- Safe Navigation through ice
- Wide swath not a problem
- Can’t measure ice thickness but tone/texture can be used to determine old from new ice (old more dangerous to navigate)
- Colour code for Nav maps
- Sentinel-1 (free) or Radarsat-2 (restricted access, pay)
What other simple applications are there for SAR?
- Arctic sea ice, forests, fires, EQ’s
Speckle
- Real but noise-like process produced by imaging radars which degrades radar image quality
- Results in bright and dark (high and low) variations that masks the true BS of the surface
- Same cover type may have high and low pixel values
- Appears as grainy salt/pepper textured appearance
What causes speckle?
- System phenomenon
- Caused by constructive and destructive interference of waves returning to radar from scatterers on surface
- Constructive when wave peaks coincide, destructive when peaks and valleys coincide
- Constructive = bright (additive from peaks)
- Destructive = dark (average high and low peak/valley to nothing)
What is a major downfall of speckle?
- Reduces ability to identify targets and separate classes
- Eg. classification of land-use
What are the 2 methods of speckle removal?
- Multi-look processing
- Speckle filtering
Multi-look processing
- Done during image formation
- Independent ‘looks’ at the same scene are averaged to reduce variations due to speckle
- Reduces speckle at the expense of spatial res
- Speckle may still appear in processed SAR image
Radarsat-2, how many looks for beam modes?
- More looks for coarser resolution modes e.g. ScanSAR has 4 x 4, Fine beam has 1x1
Speckle filtering
- Spatial filter applied during image processing chain
- Loss of spatial resolution linked to spatial filter (not averaging of looks like multi-look)
What is the image processing chain?
- Calibration, speckle filter, geometric correction, display, etc.
Speckle Filtering: Kernel
- Chosen filter moves pixel by pixel through an image and changes a particular pixel’s brightness value based on function which utilizes values w/in kernel surrounding pixel
Speckle Filtering: Larger the kernel =?
- Larger kernel = smoother result
- Larger kernel = more loss of spatial detail
- User can choose filter size, but must keep this in mind
What should an ideal speckle filter keep in mind?
- Homogeneous image areas: Filter should preserve radiometric information, BS intensity, and the edges btwn different areas
- Textured image areas: Filter should preserve radiometric information and spatial signal variability (patterns of BS that relate to scene elements)
Speckle Filtering: What does averaging do?
- Causes blurring between edges and loss of real textural components
- Averaging not always effective for speckle reduction
Speckle Filtering: Adaptive filters
- Used to preserve image detail
- Reduce some speckle while accurately preserving radiometric information, edges and texture
What is the perfect way to choose a filter type?
- No single solution for which filter type or kernel size
- Experimentation needed
- Boundaries important? Then examine various algorithms relative to problem, try different kernels
Speckle Filtering: Median filter
- Center pixel in kernel is replaced w/ median value of all pixel values w/in the kernel
- Effective at suppressing speckle noise while preserving texture and edges
Speckle Filtering: Enhanced Lee Filter
- More advanced filter
- Pixels designated as homogeneous, heterogeneous, and point target by comparing local kernel Coefficient of Variation, to class cut-off values based on Cu, Cmax and number of looks, L
- CoV = (SD/Mean), Cu = 0.523/sq. rt L, Cmax = sq. rt (1 +2/L)
- Homo if CoV Cu and CoV Cmax, no filtering
- Filter replaces the pixel in the centre of a kernel w/ new value based on its designation (homo, hetero, point)
