Aerial Photography and Remote Sensing – Study Outline

digital processing

You should be able to:

1. describe how the raw data from a satellite is turned into a useful image

2. describe why digital enhancement is necessary

3. describe how digital enhancement works

4. describe how digital processing such as PCA can aid in image interpretation

5. explain the process of digital classification

A. format of satellite remote sensing data

1. Most data is a single value for each pixel, a “digital number” (DN) representing the intensity of the measured radiation for that spot on the image

2. The digital numbers must be able to represent the full range of possible values from the instrument. Most images will not use the full range and need to be processed in some way to make a useful image.

B. preprocessing

1. digital numbers are corrected for know instrument errors or bias

2. digital numbers are recalculated to change the orientation of rows and columns to match geographic grids

3. digital numbers are recalculated to match a geographic grid to correct for distortion of the image do to lens problems, the curvature of the earth or distortion due to parallax

C. digital enhancement

1. to view an image ,each digital number is assigned to a brightness level for the display.

a. if the DN can be from 0 to 255 the simplest scheme is to just say 0=black 1= display level 1 (1/256 of full brightness) 2= display level 2 (2/256 of full brightness) and so on

b. this will often produce a useless image -- too dark, too light, no contrast…

2. histograms

a. by making a chart showing how many pixels in the image have each possible DN value, you can decide on a more useful display scheme

3. many schemes may be described as some sort of stretching, assigning a larger range of display values to a smaller range of digital numbers

a. the simplest is a linear stretch. Example:

i. the image only has DN values from 10 to 100

ii. assign 10 to black display level and 100 to full brightness

iii. spread the rest of the values as needed: 11 = display level 3, 12 = display level 6, 13 = display level 9.

b. if most of the image is represented by only a few DN values but there are a few that are very low or very high, you can choose the end points of the stretch to improve the bulk of the image at the expense of the extremes. If the DN go from 10 to 100 but 95% of the image is 40 to 60 you can assign everything <40 to black everything >60 to white and then 41= 1/20^th of full bright 42 = 2/20^th of full brightness. This would be called a selected linear stretch.

c. more complicated nonlinear schemes would assign small differences to the “rare data” and spread out the common data e.g. 1-10 = display level 1, 11-20=display 2, 21-23=display 3, 24-26=display 4…

4. you might also decide to accentuate different ranges of values by giving them different colors called a “density slice” or thematic image 0=50= blue, 51-100= green 101-125 = red…

5. digital processing

D. digital processing

1. processing involves recombining the data from different bands or “colors” in the original data

2. one very useful process is called principle components analysis (pca)

a. pca looks for colors (or bands or channels) that are well correlated. When one is dark the other is usually dark, and when one is bright the other is also bright. (If the sensor collects data on blue and yellow light but the image is mostly green then blue and yellow are well correlated.)

b. A mathematical combination of these two (or more) creates a new set of DNs for this combined color labeled PCA1. (In above example PCA1 is a number telling how bright or dark green the image is)

c. a second set of numbers PCA2 is also created to show how far away a pixel is from the combination in PCA1. (In the example, a bright blue pixel, blue high and yellow low, is a long way from PCA 1 so would be high in PCA2)

d. When each of the PCA components is displayed (and possibly stretched) with a different color, subtle differences in the original scene become obvious, and may allow interpretations that were impossible before, e.g. distinguishing between species of trees or different kinds of rock. (Third example: two pixels that are fairly dark in the normal image, one green and one blue green, would be hard to tell apart, but in the PCA images they would be dark in PCA1 but the blue green would be bright in PCA2 because it was different.)

e. PCA is especially good when you don’t know exactly what you are looking for and want the computer to help you find interesting differences across an image

3. ratio

a. another way of displaying the relationship between different measurement bands or colors Is calculate the ratio of two bands color1/color2

b. ratios take advantage of known relationships such as that between IR and green light reflected by plants but not by green paint, to highlight a feature that may be invisible in the raw image.

c. If you are looking for a “needle” with known properties in a “haystack”, a ratio may be a way to make the needle stand out.

4. index

a. more complicated calculations may help In some situations, the result is called an index

b. a well know example is the Normalized Difference Vegetation Index (NDVI)

i. NDVI = (visible – IR)/(visible + IR)

ii. NDVI value indicates how much of the pixel is covered in healthy plants, often termed its “greenness”

E. digital classification

1. involves grouping all pixels that have similar color values into a single class that hopefully represents a common land use or land cover.

a. unsupervised classification allows the computer to search for repeating patterns in the data and create it’s own classes.

b. in supervised classification the image interpreter outlines “training fields” specific areas with a known land cover and them tells the computer, “find everyplace else that looks like this”

2. selection of the proper number of classes for an image and selecting representative training fields is part of the art of digital classification