The second midterm covers weeks 6 through 10 of lecture, lab and readings.    

Note: materials below are guide to what was covered in lectures and text book, but not the articles or workbook. 

As with the first midterm, when you study for the test, focus first on lecture notes.  You should also pay close attention to the text materials from your labs.  The book will once again be an important source of information to fill in gaps in your understanding.  In particular, you will want to use the book to fill in several topics that were not covered in lecture (e.g., conversion of DNs to absolute radiance values, a number of the classification topics).  In the case of the lab materials, you should understand the broad concepts, do not worry about remembering details about the software (e.g., which sequence of commands to enter in order to access the polynomial rectification procedures, although you should know what polynomial rectification is).

The following is a list of points that you should put particular emphasis into studying.  Although I try to make the list as comprehensive as possible, there may be a question or two on the test that this list doesn't cover.  By in large, however, if you have a good grasp on the definitions, facts, concepts and skills below, you will do well on the test. 

I. The Remote Sensing System (yes, we did this before, but I want you to remember this overarching framework)

• Energy Sources

  • passive

  • active

  • the electromagnetic spectrum

• Energy Movement through the Atmosphere

  • Absorption

  • Scattering

  • Reflection

  • Transmission

• Energy/Matter Interactions at Earth's Surface

  • Absorption

  • Scattering

  • Reflection

  • Transmission

• Sensor Systems and Platforms

• Data Handling Systems

• Data User

II.  The Electromagnetic Spectrum (ditto the comment above)

• Wavelengths associated with:

  • ultraviolet light

  • visible light (blue, green, and red)

  • photographic infrared (IR)

  • near IR

  • shortwave IR

  • midrange IR

  • thermal IR

  • radar

• Range and intensity of wavelengths

  • emitted from the sun

  • emitted from the earth

• be able to draw a diagram showing the two key points above, with wavelengths and names of different ranges of energy

III.  Major Components of Image Processing Covered in this Class

• Data Restoration - the "fixing" of the data

• Image Enhancement - better visualization of key features or processes

• Image Classification - making a map of categories

IV.  Data Restoration/Image Preprocessing

• Image Rectifications (geometric corrections) 

  • Sources of distortion/error

    • Systematic (be able to give some examples)

    • Non-systematic (be able to give examples)

  • Techniques for adjusting pixel locations

    • Mathematical corrections for systematic errors

      • models for earth rotation, scanner characteristics

      • camera focal adjustments

    • Approaches for  systematic/nonsystematic error

      • polynomial approaches

        • use of ground control points

        • how different order polynomials

        •  use of root mean square (RMS) error for evaluation

      • triangulation/rubber sheeting approaches (requires same input as polynomial approaches)

      • orthorectification techniques (requires DEM)

      • reasons for choosing one of three approaches for rectifying image

  • Techniques for assigning pixel values to the new locations

    • Nearest neighbor sampling

    • Bilinear interpolation

    • Cubic convolution

    • Advantages/disadvantages of each approach
      

• Radiometric Corrections (atmospheric corrections)

  • Atmospheric scattering

    • "Relative corrections" based on information from the image, i.e. relative to the image

      • Haze removal (subtract lowest value - a poor approach)

      • Contrast stretch where lowest digital value assigned to zero (better, but still weak)

      • Regression of NIR vs Visible wavelengths to determine lowest value to subtract (better, but still weak)

    • "Absolute corrections" (i.e. correction applied independent of image data)

      • physical models based on atmospheric data (e.g., MODTRAN, ATREM, etc.)

  • Solar corrections

    • Relative correction for comparing one image to another, or part of same image with different illumination

      • comparison of same target or identical targets on both images to adjust reflectance values

    • Absolute corrections

      • for variations in Earth/Sun distance 

      • for sun angle ("sun elevation" in the book)

  • Bidirectional Reflectance

    • No unique and relatively error-free solutions at this time

  • Conversion of DNs to absolute radiance values

    • When it is important to do this

    • How this is accomplished

• Equipment Distortion (included in noise category in the text book)

  • Line dropout and cosmetic "fixes"

  • Sensor miscalibration (line striping) and approaches for correcting

• Noise Removal

  • Use of moving windows (kernels) to detect and correct noise

V. Image Enhancement

• Point Operators (Radiometric Enhancements)

  • Masks (termed "gray-level thresholding" in the book)

  • Level slicing

  • Contrast stretches

    • Basic concept of how contrast stretches work

    • Some different kinds of contrast stretches

• Local Operators (Spatial Filters)

  • Concept of "convolution," which is use of moving windows called "kernels"

  • High frequency filters ("sharpens" image)

  • Low frequency filters ("blurs" image)

  • Directional and non-directional filters

• Multi-Image or Multi-Band Enhancements (know the general concepts and when one might use the techniques below)

  • True color composites

  • False color composites

  • Intensity-hue-saturation color space transformations

  • 3-D image drape

  • Band ratioing (called "spectral ratioing" in book)

    • to remove topographic illumination effects

    • vegetation indices

  • Principal components/canonical analysis

    • to reduce spectral redundancy 

    • key features that are illuminated

  • The tasseled cap transformation

  • Fourier transforms (note: book includes this as a spatial operation and it can be used this way.  But it can also be used to examine variations in frequencies across bands)
    
    

VI.  Classification

• Supervised Classification (know general concept and advantages/disadvantages of each approach) 

  • Overall approach

    • selection of spectral training sites 

      • Training sites should represent a particular spectral signature.  One feature (e.g. water) may have several different spectral signatures (e.g., turbid/clear; shallow/deep)

      • preferably the training sites should be scattered across the image

      • training sites should be "pure" and not represent a mix of spectral signatures

      • number of training pixels per category should be a minimum of 10 times total number of categories (but preferably higher) 

      • spectral signature of training sites can come from:

        • imagery (top down approach)

        • ground based spectral measurements 

    • training site refinement

      • use of histograms and coincident spectral plots

      • quantitative indices (e.g. transformed divergence, Jeffries-Matusita distance)

      • ability of training pixels to correctly classify themselves (self classification)

  • Algorithms for classification

    • Minimum distance

    • Parallelepiped

    • Gaussian maximum likelihood classifier

• Unsupervised Classification

  • Cluster concept

  • Use of unsupervised classification:

    • to assess range of land cover in the image

    • as a guide to selecting training sites for subsequent supervised classification

    • to provide insights into range of spectral variations and their spatial distributions and extents (to see the landscape with a new eye)

• Post-Classification

  • Combining classes

  • Majority filters

  • Multi-image refinem,ents (e.g., adding a DEM)\

• Mixed Pixel Classification

  • Linear mixing/unmixing

  • Fuzzy classifications

• Accuracy Assessment

  • Error matrix (also called a confusion matrix or a contingency table)

    • Producer's accuracy

    • User's accuracy

    • Error of omission

    • Error of commission

    • KHAT statistic

  • Development of validation data sets to create error matrix

    • Validation data should not be same pixels as training sites

    • Minimum 50 validation sites (could be pixels), more if large area or many categories

    • Validation data should be from around image, not just one location (e.g., 50 pixels from one site is not appropriate)

VI.  Hyperspectral Imagery

• How hyperspectral differs from multispectral

• Applications

  • Spectral matching to spectral libraries

  • Pixel unmixing (spectral mixing)

  • Spectral angle mappings as a classifier

VII. Active Systems (Chapter 7)

• Principles of Active Radar

  • Radar = Radio Detection and Ranging

  • Wavelengths

    • K band, 0.8-2.4 cm

    • X band, 2.4-3.8 cm

    • C band, 3.8-7.5 cm

    • S band, 7.5-15 cm

    • L band, 15-30 cm

  • Major types of radar instruments

    • Plan Position Indicator (PPI) at airports, military, weather installations

    • Side Looking Radar

    • Synthetic Aperture Radar

  • Burst of single wavelength with echo/backscatter recorded

    • length of time for signal to return determines distance

    • strength of returning signal indicates composition

  • Some terms to know:

    • depression angle (angle of signal from horizontal)

    • look angle (angle of signal from nadir)

• Applications

  • Penetration of smoke, canopy, clouds, snow

  • Creation of DEMs using interferometric radar

  • Estimates of moisture content with passive microwave

  • Detection of vertically/horizontally aligned objects (e.g., trees, downed wood, rail road tracks, power lines) with polarized radar

  • and more, limited only by one's imagination

• Controls on Radar Return Signal

  • Surface Roughness relative to Wave Length

    • roughness elements < (wavelength/~25), then specular

    • (wavelength/25) < roughness element size < (wavelength/4.4), then intermediate

    • roughness > (wavelength/4.4), then diffuse

    • or in other words, smooth surfaces more mirror like, rough surfaces more diffuse

  • Depression angle

    • steeper depression angle = stronger return

  • Geometry of object

    • corner reflectors = very bright return

  • Electrical characteristics of object

    • greater conductivity = greater reflectance

    • water and metal are highly conductive 

  • Topography

    • shadows

    • layover effects 

  Spatial Resolution

  • Across Track or Range Resolution

    • Resolution improves with decreasing depression angle (i.e., with increasing distance from aircraft)

  • Along Track or Azimuth Resolution

    • Resolution improves with increasing depression angle (i.e., closer to aircraft)

• Concept of Polarization

  • Signals (HH, VV, HV, VH)

  • Generally used to detect object oriented in same way as signal

    • e.g., VV good for trees poles

    • e.g., HH good for downed wood, roads, railroads

VIII. Lidar (light detection and ranging)

• Basic concept

  • Coherent vs. normal light

  • How laser profile is created

  • small and large footprint lidar

• Multiple returns and different layers

  • First (primary); secondary and last returns (partial returns)

  • Bare earth DEM

  • Canopy mappig

  • Postings and interpolation to generate DEM

Return to Geog 4/518 Course Page