1. preparation
1.1. Basics of remote sensing:
electromagnetic spectrum, wave model and particle model, source of EMR, EMR and matter interactions, path, spectral reflectance and albedo, spectral emissivity and temperature;
remote sensing platforms;
detector configurations, dwell time, FOV and IFOV;
resolutions;
passive remote sensing (Sun, Earth, or atmosphere): multi-spectral, hyperspectral, thermal, microwave;
active remote sensing (sensor): Radar and Lidar;
digital image data formats (BSQ, BIL, BIP)
remote sensing applications (land, ocean, and atmosphere)
1.2. IDL
a programming language (command line and development environment), array-oriented, build-in functions, less use loops, zeroed array (*arr()) and index array (*indgen()), array indexing (myarr[]), array properties (size(), min(), max(), mean(), variance(), stddev(), total()), ...
plot, tv, surface, contour, itool (iplot, iimage, isurface, icontour).
1.3. Image quality and statistical
errors from the environment, random or systematic malfunction, improper analog-to-digital conversion,
histogram, univariate statistics (min, ..., ) and multivariate statistics (covariance, correlation).
2. Procedures of image processing
preprocessing, information enhancement, information extraction (image classification), accuracy assessment, post-classification.
2.1 preprocessing: radiometric correction (improving the accuracy of measurements in spectral reflectance, emittance, or back-scattered)
Internal (detector) error correction, External (atmospheric and topographic) error corrections.
absolute atmospheric correction (DN to scaled surface reflectance): radiative transfer-based, empirical line calibration.
relative radiometric correction: single-image normalization using histogram adjustment (dark subtraction), multiple-data image normalization using regression (select a base image and pseudo-invariant features-PIFs), others (flat field calibration, internal average)
topographic error correction (slope and aspect introduce radiometric distortion), image acquisition geometry (Sun's zenith angle and incidence angle, Sun and Sensor's elevation angle and azimuth angle), the key point is to calculate the Sun incidence angle.
2.2 preprocessing: geometric correction (improving the accuracy of measurement locations)
Internal geometric errors (sensor system, earth rotation, and curvature characteristics) and external geometric errors (aircraft or spacecraft's altitude and attitude changes)
geocoding and registration (image to image, image to map)
spatial interpolations (RST, Polynomial, Triangulation, difference? ) and intensity interpolations or resampling (NN, Bilinear, Cubic, difference?), ground control point, output-to-input (inverse) mapping logic, RMS error
2.3 information (image) enhancement
image reduction, image magnification, spatial and spectral profiles, contrast enhancement or stretch (linear and nonlinear), band ratioing,
linear stretch includes min-max, percentage, standard deviation, piecewise, and stretch on demands; nonlinear stretch: histogram equalization (reduce the contrast in the very light or dark parts), spatial filtering (convolution and Fourier transform), PCA, texture transformation, image sharpening
2.4 Per-pixel based image classification
unsupervised (clusters then be assigned to classes) and supervised (classes based on training sites), hard and fuzzy
K-mean and ISODATA,
parallepiped, minimum distance, mahalanobis distance, maximum likelihood (training class should be normally distributed).
what are land cover and land use?
2.5 Accuracy assessment
classification errors come from 5 sources, accuracy assessment based on training pixels or reference pixels, sample size and sampling design (as random as possible), error matrix (producer's accuracy and user's accuracy, overall (deterministic) accuracy and (deterministic) Khat accuracy), fuzzy error matrix (producer's fuzzy accuracy and user's fuzzy accuracy, overall fuzzy accuracy), GIS layers help for classification.
2.6 Post-classification and GIS
Pixel-based classification creates "salt-and-pepper", majority/minority, clump, sieve, combine, classes to vector for GIS use,
morphology filters (kernels): dilate, erode, open (erosion + dilation), close (dilation + erosion).
3. Advanced image processing
3.1 Object-oriented classification
advantages of object-oriented compared with per-pixel based method, image segmentation, criteria for segmentation (scale, color, shape, pixel neighborhood function), classification of objects (nearest neighbor and membership function), combine image with other datasets for segmentation and classification.
3.2 Hyperspectral image processing
difference between multi- and hyper- spectral imagery, endmember and pure pixel, spectral mixing, spectrum continuum and removal
steps for finding endmembers: MNF, PPI, nDV, SA; spectral hourglass
special classification methods: per-pixel (hard) and sub-pixel (fuzzy), per-pixel methods include SAM and SFF, sub-pixel methods include complete linear spectral unmixing, matched filtering, mixture-tuned matched filtering (MTMF); Tetracorder
3.3 Artificial Intelligence (AI)
Artificial neural network (ANN), decision tree, support vector machines (SVMs)