Final Exam Review Flashcards
Have cards for Lectures 6 and 7 so far (content based on the summary slides)
What general wavelength in nanometers represents the Blue electromagnetic spectrum?
400-500 nm
The purity of vividness of a color is refered to as ____________
Saturation
What are the two main applications for IHS Transformations?
- Create vivid, more distinct colours
a) Modify saturation - Merge images (fusion) from different sensors
a) e.g. combine a Landsat TM image (30m resolution) with a Landsat panchromatic (15 m resolution) image.
What is the purpose of image classification?
To Categorize pixel values into land use land cover (LULC) classes
What is the result of image classification?
A Thematic Map
what are the four factors you should consider when choosing a classification scheme?
1) Establish the intended application of the classified data
2) Define the appropriate scale for the intended application
3) Determine available and applicable (scale) remote sensing data
4) Computational resources available (hardware and software)
How many classes are in BC’s Landcover Classification?
16
What are 3 types of merges using IHS Transformations?
- Pseudo colour fusion of imagery data with ancillary data
(e.g. 6m SAR with 100m magnetic survey) - Pseudo colour fusion of two characteristics from one dataset
(e.g. 32R source elevation data with an 8U shaded relief model) - Colour transformation of two images with varying spatial resolution
(e.g. 30m Landsat 7 RGB with 15m Landsat 7 pan)
What are the 8 points of the color cube?
- Cyan
- Magenta
- Red
- Yellow
- Green
- Blue
- Black
- White
Define supervised classification
classes of the dataset are known (training data) and this know info is used to predict the classes of unknown data
What is Training Data?
info about the location and extent of the land use land cover that the image is being classified into.
List the 6 general steps of supervised classification
1) Session Configuration
2) Collect Training Sites
3) Evaluate Training Sites
4) Refine Training Sites
5) Run Classification
6) Post Classification
What does LIDAR stand for?
LIght Detection And Ranging
what is assigned during the session configuration of supervised classification?
Specifies 1) the image bands being used for display and input, 2) the channel for training data, and 3) the channel for classified output
What are the three methods for collecting training sites?
1) Manually- using mouse to draw polygons / lines
2) Using a predefined vector layer
3) Using a raster classified layer
For LIDAR, the Elevation of the Return =
Elevation of the Return = Altitude (AGL) – Distance
What is the overall purpose of evaluating training data?
to establish if your classes are distinct and representative of what you expect.
What are the four tools used for evaluating training sites and what do they all have in common?
All methods are based on evaluating uniqueness of spectral signatures.
1) Signature Separability (matrix)
2) Scatter Plot
3) Signature Statistics (mean and std.)
4) Histogram
What is the key takeaway characteristic of the refining training sites step?
It is an iterative process
What are the three supervised classification algorithms?
1) Parallelepiped
2) Minimum Distance
3) Maximum likelihood
What are 2 delivery formats for LIDAR data?
LAS & LAZ
Current version of LAS is 1.4
What are the diferences between a DSM and a DEM? What is a DTM?
-
DSM (Digital Surface Model) - captures both
the natural and built/artificial features of the
environment -
DEM (Digital Elevation Model) - represents the
bare-Earth surface, removing all natural and
built features -
DTM (Digital Terrain Model) - typically
augments a DEM, by including vector features
of the natural terrain, such as rivers and ridges
(breaklines)
What are the key differences between fusion and draping?
- Adds RGB and DEM together into 1 image
- Multiply image bands and DEM by scalar which
must equal 1 (e.g. RGB bands by 0.625 and
DEM by 0.375) - Then add each RGB band to the DEM = three
new “texturized” RGB bands -
Modifies the DN values (due to creating a
new image)
Permanent changes stored in the
database
Define the parallelepiped algorithm and its main advantage and disadvantage
- A “box” is created for each training area in each band with the size of the box determined by standard deviation. This “box” assigns the class.
- Overlap between classes is caused by high correlation in spectral response patterns of bands.
- Pixels outside of any box are classified as null.
- A tie breaker is run using maximum likelihood for all pixels falling in one than on box.
Advantage: Computationallty fast
Disadvantage: Poor accuracy due to high overlap
Define the minimum distance algorithm and its main advantage and disadvantage
- Calculates the mean DN for each class for each band
- Calculates the distance for each unknown pixel to each class mean
-Assigns unknown pixels to the nearest spectral mean
Advantage: Computationally fast and usually more accurate than PP
Disadvantage: Lead to false/forces classifications
Define the maximum likelihood algorithm and its main advantage and disadvantage
- A hyperellipse is created for each training area in each band with radii determined by standard deviation units of each class
-These hyperellipses created the classes
-The likelihood of unknown pixel being in each class is calculated and each pixel is assigned to the statistically most likely class
Advantages: Most accurate, no overlap concern
Disadvantages: Slowest computationally
Why would you want to use a null class option for the maximum likelihood algorithm?
Allows for identification and adding of classes you left out in your initial class definitions
list and describe the three post-classification processes in supervised classification
- Clean up small areas
-> use filtering to remove single-pixel classes that don’t
make sense - Generate Area Reports
-> Calculate areas for each class - Convert file formats
-> Create Shapefile or vector
What are 3 generalized categories for affecting the amount of backscattered energy from a satellite?
- Radar Viewing and Target Geometry
- Target Roughness
- Target Moisture Content
Increasing the moisture content of an object will do what to the objects diaelectric constant and how does this impact reflectivity?
- Increased moisture of an object = increased dielectric constant
- Increased dielectric constant = increased radar reflectivity
- E.g. wet soil generally appears brighter than dry soil
The term defining significant contributor to noise in a RADAR image.
Speckle
What question is central to the supervised classification workflow?
Satisfactory Results?
reminder to look at the workflow chart and be comfy filling in each step
when should you use Unsupervised classification?
- If you have little knowledge of the area
- Allows you to get a ‘feel’ for classes you expect to be in the image
- Find areas that are a class the operator had not considered
Compare and contrast classes vs clusters
Clusters are statistically separate groupings of pixels that are determined by software and algorithms. An operator can ‘seed’ the clusters with signatures.
Classes are user-defined groupings of pixels identified with training sites and that have associated statistics
List the three unsupervised classification algorithms?
K-means, ISODATA, and Fuzzy K-Means
what are the key characteristics, advantages, and drawbacks of the K-Means algorithm
Creates clusters based on spectral distance.
Assumes that the number of clusters is known prior (user-defined)
Fast and simple technique but it assumes data is normally distributed.
what are the key characteristics, advantages, and drawbacks of the ISODATA algorithm
Same processing as K-Means but also splits, merges and deletes clusters after each pass through the image.
-Allows for the number of clusters to differ from the initial value defined -> final number of clusters is defined by the algorithm.
iterations can be slow but doesn’t rely on normally distributed data
what are the key characteristics, advantages, and drawbacks of the Fuzzy K-Means algorithm
-A direct generalization of k-means
- Pixels have a gradient in membership values for each cluster
- handles continuous data, gradual boundaries, and overlapping clusters
Which unsupervised classifications are best suited for discrete data? Continuous data?
discrete = k-means and ISODATA
continuous = fuzzy k-means
Which unsupervised classifications determine the number of clusters through the algorithm itself? Which stays within the user-defined number of classes?
- ISODATA algorithm has ‘the final say’ in number of classes
- K-Means & Fuzzy K-Means don’t alter the user defined number of classes
What are the 5 steps of unsupervised classification?
1) Session Configuration
2) Algorithm Setup
3) Clustering
4) Class Creation-> a) Aggregation, b) Class Labelling
5) Post Classification
Describe the Session Configuration process of unsupervised classification
1) Add empty channels for results,
2) Define names of session, display, input bands, and output channel
Describe Algorithm setup during unsupervised classification.
Define input parameters - number of iterations and clusters
Describe the clustering process during unsupervised classification
a) Creates new image
b) generates statistical clusters
c) assigns default name “Class-xx”
d) report generated
Describe the class creation process during unsupervised classification
a) Aggregation- user groups clusters into classes
b) Class labelling- user determines the class names and finalizes the classes from the clusters
Describe the post-classification process during unsupervised classification
Apply filters
Produce area reports
Conversion to vectors
where does RADAR fall on the EM spectrum?
1mm to 1 m wavelengths
list the 5 key characteristics of active microwave sensors
1) All-weather imaging capability
2) Day and night data acquisition
3) Provides unique info about objects
-> sensitivity to targets geometry, surface roughness and moisture content
4) Can detect changes in signal characteristics
5) Provides subsurface as well as surface information
-> partial penetration of soil and vegetation canopy
what does RADAR stand for?
RAdio Detection And Ranging
Describe the principles of RADAR measurments
- The sensor transmits a microwave (radio) signal towards a target and detects the backscattered radiation. The strength of the backscattered signal is measured to discriminate between different targets and the time delay ( = 2r/c) between the transmitted and reflected signals determines the distance (r) to the target.
- The majority of polarimetric systems had a single antenna for both transmitting and receiving
What are the three different polarization combos od RADAR?
HH- horizontal send, horizontal recive
VV- vertical send, vertical receive
HV- Horizontal send, vertical receive, or vertical send, horizontal receive
What are the three general categories that affect the amount of backscattered energy?
1) Radar Viewing and Target Geometry
2) Target Roughness
3) Target Moisture Content
List and describe the three sub-components of Rader Viewing and Target Geometry as they related to the amount of backscattered energy
1) Local Incidence angle
- Slopes facing towards the radar have a small
local incidence angle resulting in stronger
backscatter (brighter image signal) and vice versa
2) Corner reflector
- occur when planer features (building sides)
intersect the surrounding ground at a right angle.
corner reflectors cause strong backscatter
3) Structure of the target
-Relationship between the transmitted signals
polarization and the structure of the target. Ex.
more vertically distributed COM/SA objects (ex. tree
truck) have stronger backscatter with vertically
polarized waves, while more horizontally disputed
COM/SA objects (ex. deciduous tree crown)will
return stronger backscatter from horizontally
polarized signals.
List and describe the three sub-components of Target Roughness as they related to the amount of backscattered energy
1) Smooth Surface
- Height variations < 1/8 of the radar wavelength
- Act like a mirror reflecting very little backscatter->
appear dark in the image (ex. calm clear water)
- No depolarization- HH and VV
2) Rough Surface
- Height variations >1/3 of the radar wavelength
- scatters the radar energy in multiple directions
- brightness increases with increasing surface
roughness
- minimal depolarization- HH and HV
3) Volume Scattering
- Occurs within vegetated surfaces
- Can cause brighter or darker appearances on
images depending on the volume interactions
- Strong depolarization - HV is very sensitive to veg.
parameters
What surface characteristics (moisture and texture) allow radar to image below the surface?
very dry and smooth
What are the wavelengths the following RADAR bands (in cm) C, L, X, P, S
X: 2.4-3.8
C: 3.8-7.5
S: 7.5-15
L: 15-30
P: 75-133
Describe the relationship between radar wavelength and vegetation structure
- Shorter wavelengths, X and C bands, interact primarily with the uppermost leaf canopy
- Longer wavelengths, L and P bands, penetrate deeper to interact with stems of smaller plants and branches and twigs of larger plants
- The longer the wavelength, the greater the sensitivity to the vertical structure of vegetation
Applications of C-band (Radarsat-2 / ERS-1 SAR) vs L-band (JERS-1’s SAR)?
C-band -> Signals backscatter significantly from:
- Small surface vegetation, dirt clods, leaves
- Small surface disturbances on the ocean
- Any object on the order of several centimeters
L-band -> backscatters in response to;
- tree trunks and larger branches
- underlying geology
- larger ocean waves
What two components control the spatial resolution of RADAR?
1) Pulse Length
2) Antenna beam width
Larger pixel size (coarser spatial resolution) is associated with longer pulse lengths and broader swaths.
Define radiometric resolution for RADAR images
the measured signal-to-noise ratio
What is the major source of noise in a RADAR image?
Speckle- obscures edges and image details, making it difficult to interpret
What are the two methods for reducing speckle and their similarities / differences?
1) Multi-look Processing
- Averages small groups of adjacent range lines into
wider multi-look range lines.
- A: reduces speckle and improves image geometry
- D: degrades the spatial resolution
2) Spatial Filtering
- A moving kernel filter that preserves the high-
frequency features (edges
-A: reduces speckle
- D: degrades the spatial resolution
Similarities: both remove speckle and degrade spatial resolution
Differences: multi-look processing is often done during data acquisition, while spatial filtering is performed on the output image in RS imagery software
What is the Temporal Resolution of RADARSAT-2 and RCM for Coherent Change Detections?
RADARSAT-2: 24 day repeat cycle
RCM: 4 day repeat cycle
List some Applications of RADAR remote sensing
1) Agriculture- crop ID, soil and crop moisture measurements
2) Forestry: Biomass estimation, fire scar mapping
3) Geology- geologic mapping
4) Hydrology- monitoring of wetlands and snow cover
5) Oceanography: sea ice ID, wave slope measurements
6) Shipping: ship detection and classification
What are the two broad categories of LiDAR and their corresponding sub types?
1) Airbone
a) Topographic- derive DEMs
b) Bathymetirc- marine terrain elevations and water
depth
2) Terrestrial
a) Mobile- car mounted
b) Static- tripod mounted
What are the typical wavelengths used by terrestrial and topographic lidar? Bathymetric LiDAR?
Terrestrial and Topo = NIR Laser as 1064nm
Bathymetric = green light at 532nm
How is the elevation of a LiDAR return calculated?
Elevation = Altitude of laser - Distance pulse travelled
Why can return numbers not be automatically fit into land type classes
Because the quantity and sequence of returns vary on the environment. In a barren area, first return could represent the ground, but in a rainforest, it would represent high vegetation.
Thus points from multiple returns are processed to determine what material they bounced off then placed into standardized classes
What are the similarities and differences between LAS and LAZ
Both are vector formats with high interoperability.
LAZ is a compressed version of LAS
LAs is the industry standard for LiDAR data storage
What information (8 things) is stored with each point within a LAS dataset?
- Number of Returns
- Return Number
- Intensity
- x, y, z values
- Scan direction
- Classification
- Scan Angle Rank
- GPS Time
What are some advantages and disadvantages of LiDAR?
Advantages
- highly accurate elevation data
- high sample density
- Penetrative abilities with multiple returns allow for
under-canopy collection
- Day and night
- No cloud shadows - acquisition in below cloud cover
Disadvantages
- Very expensive
- Indiscriminate and randomly places a LOT of x,y,z
points
- Not imagery- must be shaded or colored
- Refraction of laser pulses during heavy rain, fog, or
over water
- Low and slow operations compared to aerial
photogrammetry
- Requires skilled data analysts
What are some applications of LiDAR?
LiDAR can become cost-effective for large projects
- Right-of-way studies (oil and gas, electrical)
- Flood mapping
- Construction (volumetric analysis)
- Municipal mapping projects (transportation, planning, etc.)
- Agriculture (measure biomass, heights, volumes)
- Forestry (Vertical structure -DBH, crown cover)
- Erosion monitoring and slope stability
What is an orthophoto and it’s 3 key features?
a geometrically corrected and georeferenced aerial image.
1) Orthogonal Projection
2) Uniform Scale
3) No relief displacement
Define contours, slope, and aspect
Contours are lines representing a constant elevation above or below a datum plane.
Slope is the steepness at each cell of a raster surface (rise over run) a 100% slope is elevation gain equal to ground distance.
Aspect is the compass direction of downhill slope face
what is the wavelength range of green?
500-600nm
what is the wavelength range of red?
600-700 nm
Discuss additive versus subtractive color model.
In the additive model, the combination of all primary colours give you white/light while black is the absence of light. The subtractive model works by partially or entirely masking colours on a white background.
List and describe the key characteristics of the three computer colour models?
- CMYK (Cyan, Magenta, Yellow, Key)
-Used in print
- Subtractive model
- full combo of CMY = black
-RGB
- Combo of the primary colours
- Additive model
- white is (255, 255, 255) black is (0,0,0)
- IHS (Intensity, Hue, Saturation)
- additive model (240, 239, 240)
What/ where are the axes of the IHS cone?
Saturation (S): radius of the circular face (increasing outwards)
Hue (H): perimeter of the circular face
Intensity (I): tip of cone to circular face (increasing towards face)
Define Intensity in terms of IHS
The overall brightness of the colour
vales vary from 0% black to 100% (white), rescaled to 0 to 240
Define Hue in terms of IHS
The dominant wavelength of colour
Why would you want to merge different datasets together?
Different sensors and satellites have unique properties. By merging data from two sensors together, you can create an image that highlights the unique properties from both sensors
