click below
click below
Normal Size Small Size show me how
GIS319
FINAL FOR NR319
Question | Answer |
---|---|
Advantages of remote sensing | area covered, data format, economy |
additive color | red + blue + green = white computer screen |
subtractive color | yellow + magenta + cyan = black printer |
how are digital images different from photos? | digital is electronic image capture, not chemical digital can be numerically manipulated digital can be resized - rescaled |
Passive (vs. active) | most of the sensors - energy reflected to sensor is from another sources ie sun's energy |
Active (vs. passive) | sensor emits & recieves own energy reflected from a target ie lidar/radar |
platforms of digital remote sensing | airborne or spaceborne |
airborne (vs. spaceborne) | offers high spatial & spectral resolution. can acquire whenever it is needed |
spaceborne (vs. airborne) | sensors provide synoptic overview, often less detailed (lower spatial resolution) |
4 resolutions of remote sensing | radiometric, spatial, spectral, temporal |
radiomtric | number of separable levels of data recorded digital number - eg. 0 - 255 |
spatial | the area of ground covered by each image cell eg. 1 m vs 30 m |
spectral | the specific wavelengths caputred by a sensor |
temporal | time gap between data captures |
spatial resolution - digital images | "size" can change easily (zoom, subset, mosaic)pixel size is assumed not to change scale is referenced to physical size of the pixel ie 1 pixel =30 meters (landsat 7) |
landsat 7 has 7 bands of spectral resoluation | 3 visiable to eye - blue 1, green 2, red 3NIR 4, Mid IR 5, Thermal 6, Mid IR 7 |
Temporal Resolution near term | revisit |
Temporal resolution long term | change detection |
7 principles of photo interpretation | size, shape, shadow, tone/color, texture, patter, association/site/location |
digital number (DN) of each pixel | reflectance or brightness of each pixel |
statistical analysis | the numbers behind the image |
multispectral | pixels & digital numbers pixel (picture cell)eg. landsat 7 1 pixel = 30 m, DN = 0 - 255 |
Some important considerations for digital images (1) | Digital images are symmetrical raster datasets consisting of rows (i)& columns (j) that define "n" number of pixels |
Some important considerations for digital images (2) | multispectral images, by their nature, are compressed of more than one band (k) |
Some important considerations for digital images (3) | multiple bands are geometrically "registered" to each other... meaning they overlay each other correctly (pixels aligned) |
false color composites | we only have 3 bands of screen color to display potentially many more bands of information some of which are invisible to the naked eye |
passive optical sensors | spectral vs spatial resolution |
Ikonos | high spatial (4m) low spectral (4 bands) |
landsat MSS | moderate spatial (80m) low spectral 5 bands |
MODIS | low spatial (250-1000m)high spectral 32 bands |
Passive Optical Satellite Platform sensors | Multispectral moderate resolution sensors (landsat)Multispectral Low Spatial Resolution Sensors (AVHRR, MODIS)Multispectral high-spatial resolution sensors (hyperspatial) (IKONOS, Quickbird) |
Landsat info | 1st depolyed in 1972, Currently L5 & L7 in orbit, L7 has 7 spectral bands, 30m pixels (therma 60m, panchromatic 15m) 16 day visit (8 day by L5 or L7) landsat data continuity mission planning for next Landsat, Free imagery |
MODIS | multispectral/multispatial resolution sensor w/ 2 satellites - Aqua & terraGlobal coverage 2x daily |
MODIS uses | surface reflectance, land surface temperature & emissivity, land cover/change, vegetation indices, thermal anomalies/fire |
Image processing | file formats |
Digital remote sensing introduce bands of inormation represented as | digital numbers per pixel |
digital reomte sensing primarily uses | digital data from spaceborne platforms |
we can numberically process digital images in many way | beyond basic visual interpretation |
Landsat program has been the workhorse of remote sensing with | other platforms/sensors providing information at differnt resolutions |
two types of classiciation | supervised & unsupervised |
image enhancement | basic image manipulation |
convolution (neighborhood analysis | involves the passing of a moving window over an image & creating a new image where each pixel in the new image is a function of the original pixel values within the moving window & the coefficents of the moving window as specified by the user |
convolution kernel | a windowmay be considered as matrix (or mask) of coefficients that are to be multiplied by image pixel values to derive a new pixel value for a resultant ehanced image. This matrix may be of any size in pixels & does not necessarily have to be square. |
image enhancement convolution | the shape of the kernel is applied to te image to create a target neighborhoodsum(kernel x image neighborhood)/sum(kernel) |
Spatial filters low pass | a low pas (mean) filter tends to generate the image) |
spatial filters edge | identify gradients/transitions between pixel values such as faults, road cuts outcrops |
image enhancement image ratios (1) | healthy vegetation reflects strongly in the near-infrared portion of the spectrum while absorbing strongly in the visible red. |
image enhancement image ratios (2) | other surface types, such as soil & wter, show near equal reflectances in both the enar-infrared & red portions |
image enhancement image ratios (3) | thus the discrimination of vegetation from other surface cover types in significantly enhanced |
image enhancementvegetation indices/ ratios | SVI = NIR/red NDVI = NIR-red/NIR+red NBR=NIR-MIR(7)/NIR+MIR(7) |
SVI | SIMPLE VEGETATION INDEX |
NDVI | normalized difference vegetation index |
image classification | matching spectral classs to information classes |
image classification definition: | turning data into information |
defined image classification: (1) | image classification uses the spectral information represented by the digital numbers in one or more spectral bands, and attempts to classify each individual pixel based on this spectral information |
defined image classification: (2) | the objective is to assign all pixels in the image to particular class or themes ( eg water, coniferous forest, deciduous forest, corn, wheat,e tc |
defined image classification: (3) | the resulting classified image is comprised of a mosaic of pixels, each of which belong to a particular theme, and is a thematic map |
defined image classification: (4) | classification can then be tested for accuracy |
defined image classification: (5) | information classes are those categories of interest that the analyst is actually trying to identify in the imagery or at least has field data on |
example of image classes field data | different kinds of crops, different forest types or tree species, differnt geoligical units or rock types, etc. |
defined image classification: (6) | spectral classes are groups of pixels that are uniform (or near-similar) with respect to their brightness values in the image - they tend to cluster around a mean value |
defined image classification: (7) | the objective is to match the spctral classes in the data to the imformation classes of intrest |
defined image classification: (8) | rarely do spectral classes perfectly match information classes |
defined image classification: (9) | a broad information class (eg forest) may contain a number of spectral sub classes with unique spectral variations |
Feature space | plotting reflectance in various wavlengths |
Image classification - supervised | select training fields, evaluate signatures, classify image, evaluate classification |
image classification - unsupervised | run cluster algorithm, classify image, evaluate signatures, evaluate classification |
Unsupervised classification (1) | litterally allowing the comupter to identify statistically similar classes of pixel values=clusters |
Unsupervised classification (2) | does not require prior inforamtion (a priori) of the subject area - meaning you can classify without field data |
Unsupervised classification (3) | you can still have to compare your classification with reality eventually |
Unsupervised classification (4) | a classification scheme must be in place - are you a lumper or splitter? |
Unsupervised classification (5) | you set the number of classes - the computer calculates the appropriate clusters |
Lumping vs splitting too few | numerous cover types sharing spectral classes, |
Lumping vs splitting too many | redunant classes that need to be combined |
Lumping vs splitting | either too many or too few you will need to combine (lump) some classes & break apart (split) others |
Unsupervised classification (6) | no prior knowledge but there needs to be some knoledge or the are to interpret the resulting classes |
Unsupervised classification (7) | the opportunity for input error is minimized |
Unsupervised classification (8) | unique classes are recognised as distinct clusters of unique spectral classes |
Unsupervised classification (9) | challenge is to convert these data classes into accurate information classes |
Unsupervised classification challenges (1) | classes are statistically based - may not match field data well |
Unsupervised classification challenges (2) | spectral classes are not necessarily information classes |
Unsupervised classification challenges (3) | you still have to interpret the classes & decide whether to lump or split classes |
supervised classification (1) | uses representative training sites to direct or train the computer to identify all similar pixels and therefore classes |
supervised classification (2) | requires prior knowledge to determine training sites |
supervised classification (3) | classification algorithum/rule (ie maximum likelihood) is chosen and training sites applied |
supervised classification (4) | results are compared to known field data & accuracy assessed |
four main stages involved in supervised classification are: | training stage, classification stage, output stage, accuracy assessment stage |
image enhancement summary I (1) | digital imagery allows a wide variety of enhancements to be applied |
image enhancement summary I (2) | "simple" acts of zooming, subsetting, filtering provide improved image analysis - compare that to traditional photo |
image enhancement summary I (3) | classification is the process of organizing the pixels of an image into similar value classes |
image enhancement summary I (4) | the key to classification is turning pixel values into information |
image enhancement summary I (5) | the 2 fundamental types of classification are unsupervised & supervised |
image enhancement summary I (6) | unsupervised requires no prior knowledge of an area and its cover types |
image enhancement summary I (7) | supervised requries prior knowledge through the use of training areas |
image enhancement summary I (8) | remember, remote sensing does not occur in a vacuum - you have to bring knowldge & experience to the analysis to be effective & accurate |
image enhancement summary II (1) | remote sensing allows the acquisition of information about a place through interpretation & measurment of images |
image enhancement summary II (2) | GIS & remote sensing are inherently tied together |